CN112966810A - Helmet detection method and device based on improved YOLOv5s, electronic equipment and storage medium - Google Patents
Helmet detection method and device based on improved YOLOv5s, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN112966810A CN112966810A CN202110144067.2A CN202110144067A CN112966810A CN 112966810 A CN112966810 A CN 112966810A CN 202110144067 A CN202110144067 A CN 202110144067A CN 112966810 A CN112966810 A CN 112966810A
- Authority
- CN
- China
- Prior art keywords
- yolov5s
- helmet
- image
- target detection
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 197
- 238000012549 training Methods 0.000 claims abstract description 44
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 20
- 238000000034 method Methods 0.000 claims abstract description 13
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 27
- 125000000205 L-threonino group Chemical group [H]OC(=O)[C@@]([H])(N([H])[*])[C@](C([H])([H])[H])([H])O[H] 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 22
- 230000004913 activation Effects 0.000 claims description 13
- 238000004891 communication Methods 0.000 claims description 12
- 230000009466 transformation Effects 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 230000010365 information processing Effects 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 12
- 230000000694 effects Effects 0.000 description 6
- 206010019196 Head injury Diseases 0.000 description 3
- 206010039203 Road traffic accident Diseases 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
本发明涉及一种基于改进YOLOv5s的头盔检测方法、装置、电子设备和存储介质，本方法包括：构建YOLOv5s‑Light目标检测框架；利用目标图像训练集对所述YOLOv5s‑Light目标检测框架进行训练得到YOLOv5s‑Light目标检测训练框架；将待检测图像集输入所述YOLOv5s‑Light目标检测训练框架中得到目标检测集合；利用骑行检测算法对所述目标检测集合进行检测得到目标检测结果。本方法首先构建了YOLOv5s‑Light目标检测框架，YOLOv5s‑Light目标检测框架在YOLOv5s目标检测框架的基础上大幅度的减少了网络参数量，实现了轻量化模型的构建，然后利用YOLOv5s‑Light目标检测框架得到目标检测集合，最后通过骑行检测算法对目标检测集合进行检测得到目标检测结果，能够有效检测需求的目标，过滤不必要的目标，检测要求低。
The invention relates to a helmet detection method, device, electronic device and storage medium based on improved YOLOv5s. The method includes: constructing a YOLOv5s-Light target detection frame; using a target image training set to train the YOLOv5s-Light target detection frame to obtain YOLOv5s-Light target detection training framework; input the image set to be detected into the YOLOv5s-Light target detection training framework to obtain a target detection set; use a riding detection algorithm to detect the target detection set to obtain a target detection result. This method firstly constructs the YOLOv5s‑Light target detection framework. The YOLOv5s‑Light target detection framework greatly reduces the amount of network parameters based on the YOLOv5s target detection framework, and realizes the construction of a lightweight model, and then uses the YOLOv5s‑Light target detection framework. The framework obtains the target detection set, and finally detects the target detection set through the riding detection algorithm to obtain the target detection result, which can effectively detect the required targets, filter unnecessary targets, and have low detection requirements.
Description
Technical Field
The invention belongs to the technical field of machine vision application, and particularly relates to a helmet detection method and device based on improved YOLOv5s, electronic equipment and a storage medium.
Background
In recent years, as important transportation means for people to go out daily, traffic accidents caused by motorcycles and electric vehicles are increasing, the death rate caused by head injury after the traffic accidents happen to riders of the motorcycles and the electric vehicles is very high, about 80 percent of the riders are craniocerebral injuries, and the important reason for the situation is that the riders cannot wear helmets correctly and neglect head protection.
Scientific studies have shown that if a safety helmet is worn correctly, the helmet can absorb most of the impact energy when an accident occurs, reducing head injury caused by the accident, thereby reducing the risk of traffic accident death by 60% to 70%. At present, people do not pay enough attention to the problem of correctly wearing the helmet, the traditional helmet wearing detection method depends on the supervision of a traffic police or manual work to detect the content shot by the camera equipment, and the traditional helmet wearing detection method is low in efficiency and high in cost. Therefore, the design of an automatic helmet wearing condition detection method is particularly important for improving the helmet detection precision and efficiency, assisting traffic police law enforcement and realizing intelligent traffic management. The existing helmet detection methods are mainly divided into two categories, namely a helmet detection method based on traditional machine learning and a helmet detection method based on deep learning. The detection precision and efficiency of the traditional machine learning-based helmet detection method are required to be further improved. For example, Bhaga et al propose a helmet detection method based on Local Binary Pattern (LBP) cascade classifier, but the training and detection process is very time-consuming.
In order to optimize the helmet detection method of the traditional machine learning, muktar et al first extract features of a motorcycle by using a Histogram of Oriented Gradients (HOG) feature extraction algorithm, and then classify and detect the helmet by using a Support Vector Machine (SVM). In the helmet detection method based on deep learning, Prajwal et al firstly use YOLOv2 to detect people, motorcycles and electric vehicles in video frames, and then use YOLOv3 to detect whether a Region of interest (ROI) contains a helmet or not, and the method uses frame redundancy and has large calculation amount; noel et al first classify motorcycles and electric vehicles using conventional machine vision methods, and then detect helmets using a Yolov3 target detection framework, which has high requirements for shooting angle and image quality.
Disclosure of Invention
In order to solve the above problems in the prior art, the present invention provides a helmet detection method, device, electronic device and storage medium based on improved YOLOv5 s. The technical problem to be solved by the invention is realized by the following technical scheme:
the invention relates to a helmet detection method based on improved YOLOv5s, which comprises the following steps:
constructing a YOLOv5s-Light target detection framework;
training the YOLOv5s-Light target detection frame by using a target image training set to obtain a YOLOv5s-Light target detection training frame;
inputting the image set to be detected into the YOLOv5s-Light target detection training frame to obtain a target detection set;
and detecting the target detection set by using a riding detection algorithm to obtain a target detection result.
In one embodiment of the invention, the YOLOv5s-Light target detection framework is constructed, and comprises the following steps:
based on the idea of lightweight network, performing first convolution structure transformation on a residual block of a first YOLOv5s network to obtain a second YOLOv5s network;
replacing an activation function of a deep level residual block of the second YOLOv5s network with an H-Swish function to obtain a third YOLOv5s network;
adding an SE attention module in a residual block of the third YOLOv5s network to obtain a fourth YOLOv5s network based on a channel attention mechanism;
compressing the number of channels output by each layer of the fourth YOLOv5s network at the same ratio to obtain a fifth YOLOv5s network;
and performing second convolution structure transformation on the fifth YOLOv5s network to obtain the YOLOv5s-Light target detection framework.
In one embodiment of the invention, the first convolution structure transformation comprises:
changing the common convolution in Bottleneck of the residual block in the first YOLOv5s network into packet convolution;
the inversion of bottleeck of the residual block in the first YOLOv5s network is cancelled.
In one embodiment of the invention, the second convolution structure transform comprises:
replacing the convolution of the non-residual block Bottleneck in the fifth YOLOv5s network with a depth separable convolution.
In one embodiment of the invention, the target image training set comprises a first head target image, a first helmet-worn head image, and a first rider image.
In an embodiment of the present invention, detecting the target detection set by using a riding detection algorithm to obtain a target detection result, includes:
initializing a rider image set wearing a helmet and a rider image set not wearing the helmet;
obtaining a second head target image and/or a second helmet-free rider image by using the YOLOv5s-Light target detection framework detection test set;
using the second head target image and the second unworn helmet rider image to arrive at a best IoU threshold;
detecting the target detection set by using the YOLOv5s-Light target detection framework to obtain a third head target image, a third helmet-worn head image and a third rider image;
placing the third helmet-worn head image in the helmet-worn rider image collection;
deriving a third unworn helmet rider image using the third head target image, the third rider image and the optimal IoU threshold value;
placing the third helmet-free rider image in the helmet-free rider image set, the helmet-free rider image set and the helmet-free rider image set being the target detection results.
In one embodiment of the present invention, using the second head target image and the second unworn helmet rider image to derive an optimal IoU threshold comprises:
step 4.3.1, initializing a threshold null set and an IoU threshold, wherein the initialization value of the IoU threshold is thres;
step 4.3.2, obtaining a second rider image by utilizing the YOLOv5s-Light target detection framework detection test set;
step 4.3.3, calculating IoU thresholds of said second head target image and said second rider image resulting in a second IoU threshold set;
step 4.3.4, determining whether the IoU threshold value in the second IoU threshold value set is greater than the thres, if the IoU threshold value in the second IoU threshold value set is greater than or equal to the thres, adding the second head target image into the threshold null set, if the IoU threshold value in the second IoU threshold value set is less than the thres, ignoring the second head target image until the comparison of all IoU threshold values in the second IoU threshold value set with the thres is completed;
step 4.3.5, calculating the difference value between the number of the second head target images in the threshold empty set and the number of the second helmet-unworn rider images, and adding the difference value into a target difference value set;
step 4.3.6, after adding the difference value into a target difference value set, reducing the thres by 0.01, and emptying the threshold empty set;
and 4.3.7, repeating the steps 4.3.4-4.3.6 until the value of thres is 0, and the value of thres corresponding to the minimum difference in the target difference set is the optimal IoU threshold.
In one embodiment of the present invention, there is provided a helmet detection device based on modified YOLOv5s, comprising:
the model building module is used for building a YOLOv5s-Light target detection framework;
the model training module is used for training the YOLOv5s-Light target detection framework by using a target image training set to obtain a YOLOv5s-Light target detection training framework;
the information processing module is used for inputting the image set to be detected into the YOLOv5s-Light target detection training frame to obtain a target detection set;
and the riding detection module is used for detecting the target detection set by utilizing a riding detection algorithm to obtain a target detection result.
One embodiment of the present invention provides a helmet detection electronic device based on modified YOLOv5s, including: the system comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus;
a memory for storing a computer program;
a processor for implementing the method steps when executing the computer program.
An embodiment of the present invention provides a computer readable storage medium having stored thereon a computer program which, when being executed by a processor, implements the steps of the improved YOLOv5 s-based helmet detection method.
Compared with the prior art, the invention has the beneficial effects that:
the application discloses a helmet detection method based on improved YOLOv5s, the method comprises the steps of firstly constructing a YOLOv5s-Light target detection frame, greatly reducing the number of network parameters on the basis of the YOLOv5s target detection frame through the YOLOv5s-Light target detection frame, achieving construction of a lightweight model, then obtaining a target detection set through the YOLOv5s-Light target detection frame, and finally detecting the target detection set through a riding detection algorithm to obtain a target detection result.
Drawings
Fig. 1 is a flowchart of a helmet detection method based on improved YOLOv5s according to an embodiment of the present invention;
fig. 2 is a flowchart illustrating an operation of replacing an activation function of a deep residual block of a second YOLOv5s network with an H-Swish function according to an embodiment of the present invention;
FIG. 3a is a diagram illustrating a standard convolution operation according to an embodiment of the present invention;
FIG. 3b is a schematic diagram illustrating a depth separable convolution operation according to an embodiment of the present invention;
FIG. 4 is an effect diagram of a riding detection algorithm provided by an embodiment of the present invention;
fig. 5 is a diagram illustrating a partial helmet detection effect according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a helmet detection device based on the improved YOLOv5s according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of a helmet detection electronic device based on the improved YOLOv5s according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
It should be noted that the terms "first", "second" and "third" in the description of the present invention are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", "third" may explicitly or implicitly include one or more of the features.
Example one
Referring to fig. 1, fig. 1 is a flowchart of a helmet detection method based on improved YOLOv5s according to an embodiment of the present invention. The embodiment discloses a helmet detection method based on improved YOLOv5s, which comprises the following steps:
step 1, constructing a YOLOv5s-Light target detection framework.
Specifically, the present embodiment improves the algorithm of the first YOLOv5s network, and adopts a depth separable convolution module, an SE (Squeeze-and-excite) attention module and an improved inversion residual in the network model design, and the activation function adopts an H-swish function. The lightweight network model YOLOv5s-Light target detection framework is obtained while the detection frame rate and the accuracy are ensured, the parameter number of the model is greatly reduced, the efficiency of the model is improved, and the lightweight helmet detection method easy to deploy is realized.
Further, step 1 comprises:
step 1.1, based on the idea of lightweight network, performing first convolution structure transformation on a residual block of a first YOLOv5s network to obtain a second YOLOv5s network.
Further, the first convolution structure transformation includes:
changing the ordinary convolution in Bottleneck of the residual block in the first YOLOv5s network into packet convolution;
the inversion of bottleeck of the residual block in the first YOLOv5s network is cancelled.
Specifically, the normal convolution in the standard residual block bottleeck of the first YOLOv5s network is changed to a Group convolution (Group convolution), and the inversion of the residual block bottleeck of the first YOLOv5s network is cancelled to reduce the network parameters.
The first YOLOv5s network is an original YOLOv5s network, and is difficult to directly deploy on an edge computing platform, the detection speed is slow, the detection scene of the embodiment is actual traffic, so the requirement on the real-time performance of detection is high, the embodiment can greatly reduce network parameters under the idea of lightweight network, and the overall structure of the network is not changed.
And step 1.2, replacing the activation function of the deep level residual block of the second YOLOv5s network with an H-Swish function to obtain a third YOLOv5s network.
Referring to fig. 2, fig. 2 is an operation flowchart of replacing the activation function of the deep residual block of the second YOLOv5s network with an H-Swish function according to an embodiment of the present invention, in this embodiment, the activation function of the deep residual block of the second YOLOv5s network is replaced with an H-Swish function to obtain a third YOLOv5s network, and the activation function of the shallow residual block of the second YOLOv5s network is still maintained as LeakyReLu by encouraging the function to use fewer parameters to learn sparser image characteristics.
Further, the expression of the H-Swish function is:
ReLU represents the ReLU activation function for hidden neuron outputs, and x represents the input variables of the second YOLOv5s network.
And step 1.3, adding an SE attention module in a residual block of the third YOLOv5s network based on a channel attention mechanism to obtain a fourth YOLOv5s network.
Specifically, the SE attention module is specifically an Squeeze-And-Excite module, And an SE attention module is added to a residual block of the third YOLOv5s network, And global pooling can be used as channel weighting of the feature map, so that the feature learning capability of the network on the main target is improved.
And step 1.4, compressing the number of channels output by each layer of the fourth YOLOv5s network at the same ratio to obtain a fifth YOLOv5s network.
Specifically, the number of channels output by each layer of the fourth YOLOv5s network is compressed by the same ratio to further compress the parameter number of the fourth YOLOv5s network, and the compression ratio can be adjusted according to the hardware condition of the actually deployed edge computing platform.
And step 1.5, performing second convolution structure transformation on the fifth YOLOv5s network to obtain a YOLOv5s-Light target detection framework.
Further, the second convolution structure transform includes:
the convolution of the non-residual block bottleeck in the fifth YOLOv5s network is replaced with a depth separable convolution (Depthwise separable convolution).
Referring to fig. 3a and fig. 3b, fig. 3a is a schematic diagram of a standard convolution operation according to an embodiment of the present invention; FIG. 3b is a diagram illustrating a depth separable convolution operation according to an embodiment of the present invention.
Specifically, the convolution of the non-residual block bottleeck in the fifth YOLOv5s network is replaced by a depth separable convolution, that is, the normal convolution is split into a "convolution only, non-summation" depth convolution operation and a channel-adaptive 1 × 1 point convolution to reduce the parameters.
The network structure of the resulting improved YOLOv5s-Light target detection framework is shown in Table 1.
TABLE 1
Wherein, Conv represents the depth separable convolution operation (including the activation function and the BN layer), bneck represents the improved residual block, LRE represents the use of the Leaky-ReLu activation function, HS represents the use of the H-Swish activation function, and concat (x) represents the channel fusion of the previous layer output and the x-th layer output of the network of the YOLOv5s-Light target detection framework.
Further, the parameter quantity of the improved YOLOv5s network is reduced by 17.9 times compared with that before improvement, and the construction of a lightweight model is realized.
And 2, training the YOLOv5s-Light target detection frame by using a target image training set to obtain a YOLOv5s-Light target detection training frame.
In particular, the target image training set includes a first head target image (head), a first helmet-worn head image (helmet), and a first rider image (motorperson).
And 3, inputting the image set to be detected into a YOLOv5s-Light target detection training frame to obtain a target detection set.
Specifically, the present embodiment detects only whether or not the rider of the motorcycle or the electric vehicle wears the helmet, and does not detect other vehicles or pedestrians. The method comprises the steps of shooting scene images of the traffic intersection by using high-definition camera equipment erected at the traffic intersection, and inputting each frame of shot images serving as images to be detected into a YOLOv5s-Light target detection training frame to obtain a target detection set. The target detection set comprises a head target image, a head image of a helmet and a rider image.
The head target image represents that a head target can be detected in the image; the helmet-worn head image representation image shows that a helmet-worn head target can be detected; the rider image indicates that a rider target wearing a helmet and a rider target not wearing a helmet can be detected in the image, wherein the rider's ride-on tools are only a motorcycle and an electric vehicle.
And 4, detecting the target detection set by using a riding detection algorithm to obtain a target detection result.
Referring to fig. 4 and 5, fig. 4 is a diagram illustrating the effect of a riding detection algorithm according to an embodiment of the present invention; fig. 5 is a diagram illustrating a partial helmet detection effect according to an embodiment of the present invention.
Specifically, the head target image detected in step 3 is automatically filtered for pedestrians by a riding detection algorithm, which aims to filter out all pedestrian targets whose Intersection ratio (IoU) with the category target frame of the rider image is smaller than the adaptive threshold.
Further, step 4 comprises:
and 4.1, initializing a rider image set with a helmet and a rider image set without the helmet.
Specifically, two empty sets are initialized, one set H of images of a rider wearing a helmet and one set U of images of a rider without a helmet.
And 4.2, obtaining a second head target image and/or a second helmet-free rider image by utilizing a YOLOv5s-Light target detection framework detection test set.
Specifically, the test set is input into a YOLOv5s-Light target detection framework to be detected to obtain a second head target image and/or a second helmet-free rider image, and the second head target image and/or the second helmet-free rider image are both image sets. The images of the test set were used to obtain the optimal IoU threshold using the YOLOv5s-Light object detection framework.
Step 4.3, the optimal IoU threshold is obtained using the second head target image and the second helmet-less rider image.
Further, step 4.3 comprises:
step 4.3.1, initializing a threshold null set and IoU thresholds, wherein the initialization value of IoU threshold is thres;
specifically, an empty set T is initialized as a threshold empty set, and a threshold is initialized IoU as a variable thres, with a preferred value of thres being 1.
Step 4.3.2, obtaining a second rider image by utilizing a YOLOv5s-Light target detection framework detection test set;
specifically, the test set is input into a YOLOv5s-Light target detection framework for detection, and a plurality of second rider images are obtained.
Step 4.3.3, deriving a second IoU threshold set from the set of second head target images and the set of second rider images;
specifically, the IoU threshold value is calculated by the cross-over ratio formula, where a is the set of second head target images, and B is the set of second rider images, the second IoU threshold value set total calculation formula is:
step 4.3.4, determining whether the IoU threshold value in the second IoU threshold value set is greater than thres, if the IoU threshold value in the second IoU threshold value set is greater than or equal to thres, adding the second head target image into a threshold value empty set, if the IoU threshold value in the second IoU threshold value set is less than thres, ignoring the second head target image until the comparison of all IoU threshold values in the second IoU threshold value set with thres is completed;
4.3.5, calculating the difference value between the number of second head target images in the threshold empty set and the number of second helmet-unworn rider images, and adding the difference value into a target difference value set;
step 4.3.6, after the difference value is added into the target difference value set, reducing thres by 0.01, and emptying a threshold empty set;
and 4.3.7, repeating the steps 4.3.4-4.3.6 until the value of thres is 0, and the value of thres corresponding to the minimum difference in the target difference set is the optimal IoU threshold.
And 4.4, detecting the target detection set by using a YOLOv5s-Light target detection framework to obtain a third head target image, a third helmet-wearing head image and a third rider image.
Specifically, inputting the target detection set into YOLOv5s-Light target detection framework detects a third head target image, a third helmet-worn head image, and a third rider image, which are all image sets.
And 4.5, placing the head image of the third helmet wearing into the rider image set wearing the helmet.
Specifically, the acquired head image of the third helmet-worn subject is put into a helmet-worn rider image set to filter the detected image.
Step 4.6, a third helmet-free rider image is obtained using the third head target image, the third rider image and the optimal IoU threshold value.
Specifically, IoU thresholds for the third head target image and the third rider image are calculated, and if the IoU threshold is greater than the optimal IoU threshold, a third helmet-free rider image is obtained.
And 4.7, placing the third helmet-free rider image into a helmet-free rider image set, wherein the helmet-free rider image set and the helmet-free rider image set are combined into a target detection result.
Specifically, the acquired head image of the third helmet-worn person is placed in a helmet-worn rider image set, and the third helmet-unworn rider image is placed in a helmet-unworn rider image set, so that detection and classification of the target image are completed.
In summary, in the embodiment, the YOLOv5s-Light target detection training frame is used to perform target detection on the image set to be detected to obtain the target frame of the head target image, the head image worn on the helmet and the image of the rider, so that the number of networks is greatly reduced, the number of network parameters after improvement is reduced by 17.9 times compared with the number of network parameters before improvement, and the lightweight model is constructed. Then, the riding detection algorithm is utilized to filter the head target image, the head image wearing the helmet and the target frame of the rider image, so that the rider target of the motorcycle and the electric vehicle can be effectively detected, and filtering of other targets such as pedestrians is equivalent.
Example two
Referring to fig. 6, fig. 6 is a schematic structural diagram of a helmet detection device based on the improved YOLOv5s according to an embodiment of the present invention.
The embodiment discloses a helmet detection device based on modified YOLOv5s, including:
the model building module 1 is used for building a YOLOv5s-Light target detection framework;
the model training module 2 is used for training the YOLOv5s-Light target detection frame by using a target image training set to obtain a YOLOv5s-Light target detection training frame;
the information processing module 3 is used for inputting the image set to be detected into a YOLOv5s-Light target detection training frame to obtain a target detection set;
and the riding detection module 4 is used for detecting the target detection set by utilizing a riding detection algorithm to obtain a target detection result.
In an embodiment of the present invention, the model building module 1 is configured to train a YOLOv5s-Light target detection framework using a target image training set, where the image training set to be detected is used to obtain a target detection set.
In an embodiment of the invention, the model training module 2 constructs the YOLOv5s-Light target detection framework through methods such as a lightweight network idea, activation function replacement, a channel attention mechanism, and network channel number compression, respectively.
In an embodiment of the present invention, the information processing module 3 is configured to input the set of images to be detected into a YOLOv5s-Light target detection training framework to obtain a target detection set, where the target detection set includes a head target image, a head image wearing a helmet, and a rider image.
In an embodiment of the present invention, the riding detection module 4 is configured to detect the target detection set by using a riding detection algorithm to obtain a target detection result, where the riding detection algorithm is to filter out all pedestrian targets whose Intersection ratio (IoU) with a category target frame of the image of the rider is smaller than an adaptive threshold.
EXAMPLE III
Referring to fig. 7, fig. 7 is a schematic structural diagram of a helmet detection electronic device based on the improved YOLOv5s according to an embodiment of the present invention.
The embodiment discloses a helmet detection electronic device based on improved YOLOv5s, including: the system comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus;
a memory for storing a computer program;
a processor for implementing the method steps of any of the embodiments when executing the computer program.
The helmet detection electronic device based on the improved YOLOv5s provided by the embodiment of the present invention can implement the above method embodiments, and the implementation principle and technical effect are similar, and are not described herein again.
Example four
Yet another embodiment of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of:
constructing a YOLOv5s-Light target detection framework;
training the YOLOv5s-Light target detection frame by using a target image training set to obtain a YOLOv5s-Light target detection training frame;
inputting an image set to be detected into a YOLOv5s-Light target detection training frame to obtain a target detection set;
and detecting the target detection set by using a riding detection algorithm to obtain a target detection result.
The computer-readable storage medium provided by the embodiment of the present invention may implement the above method embodiments, and the implementation principle and technical effect are similar, which are not described herein again.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (10)
1. A helmet detection method based on improved YOLOv5s is characterized by comprising the following steps:
constructing a YOLOv5s-Light target detection framework;
training the YOLOv5s-Light target detection frame by using a target image training set to obtain a YOLOv5s-Light target detection training frame;
inputting the image set to be detected into the YOLOv5s-Light target detection training frame to obtain a target detection set;
and detecting the target detection set by using a riding detection algorithm to obtain a target detection result.
2. The helmet detection method based on the improved YOLOv5s of claim 1, wherein constructing a YOLOv5s-Light target detection framework comprises:
based on the idea of lightweight network, performing first convolution structure transformation on a residual block of a first YOLOv5s network to obtain a second YOLOv5s network;
replacing an activation function of a deep level residual block of the second YOLOv5s network with an H-Swish function to obtain a third YOLOv5s network;
adding an SE attention module in a residual block of the third YOLOv5s network to obtain a fourth YOLOv5s network based on a channel attention mechanism;
compressing the number of channels output by each layer of the fourth YOLOv5s network at the same ratio to obtain a fifth YOLOv5s network;
and performing second convolution structure transformation on the fifth YOLOv5s network to obtain the YOLOv5s-Light target detection framework.
3. The helmet detection method based on the improved YOLOv5s of claim 2, wherein the first convolution structure transformation comprises:
changing the ordinary convolution in Bottleneck of the residual block in the first YOLOv5s network into a packet convolution;
the inversion of bottleeck of the residual block in the first YOLOv5s network is cancelled.
4. The helmet detection method of claim 2 based on improved YOLOv5s, wherein the second convolution structure transformation comprises:
replacing the convolution of the non-residual block Bottleneck in the fifth YOLOv5s network with a depth separable convolution.
5. The improved YOLOv5 s-based helmet detection method of claim 1, wherein the target-image training set comprises a first head target image, a first helmet-worn head image, and a first rider image.
6. The helmet detection method based on the improved YOLOv5s of claim 1, wherein the detecting the target detection set by using a riding detection algorithm to obtain a target detection result comprises:
initializing a rider image set wearing a helmet and a rider image set not wearing the helmet;
obtaining a second head target image and/or a second helmet-free rider image using the YOLOv5s-Light target detection framework detection test set;
deriving an optimal IoU threshold using the second head target image and the second unworn helmet rider image;
detecting the target detection set by using the YOLOv5s-Light target detection framework to obtain a third head target image, a third helmet-worn head image and a third rider image;
placing the third helmet-worn head image in the helmet-worn rider image collection;
obtaining a third unworn helmet rider image using the third head target image, the third rider image, and the optimal IoU threshold;
placing the third helmet-free rider image in the helmet-free rider image set, the helmet-worn rider image set and the helmet-free rider image set being the target detection result.
7. The improved YOLOv5 s-based helmet detection method of claim 6, wherein using the second head target image and the second unworn helmet rider image to derive an optimal IoU threshold comprises:
step 4.3.1, initializing a threshold null set and an IoU threshold, wherein the initialization value of the IoU threshold is thres;
step 4.3.2, obtaining a second rider image by utilizing the YOLOv5s-Light target detection framework detection test set;
step 4.3.3, deriving a second IoU threshold set from said set of second head target images and said set of second rider images, said second IoU threshold set comprising deriving IoU thresholds from said second IoU threshold set summation calculation formula;
step 4.3.4, determining whether the IoU threshold value in the second IoU threshold value set is greater than the thres, if the IoU threshold value in the second IoU threshold value set is greater than or equal to the thres, adding the second head target image into the threshold null set, if the IoU threshold value in the second IoU threshold value set is less than the thres, ignoring the second head target image until the comparison of all IoU threshold values in the second IoU threshold value set with the thres is completed;
step 4.3.5, calculating the difference value between the number of the second head target images in the threshold empty set and the number of the second helmet-unworn rider images, and adding the difference value into a target difference value set;
step 4.3.6, after adding the difference value into a target difference value set, reducing the thres by 0.01, and emptying the threshold empty set;
and 4.3.7, repeating the steps 4.3.4-4.3.6 until the value of thres is 0, and the value of thres corresponding to the minimum difference in the target difference set is the optimal IoU threshold.
8. Helmet detection device based on modified YOLOv5s, comprising:
the model building module (1) is used for building a YOLOv5s-Light target detection framework;
a model training module (2) for training the Yolov5s-Light target detection framework by using a target image training set to obtain a Yolov5s-Light target detection training framework;
the information processing module (3) is used for inputting the image set to be detected into the YOLOv5s-Light target detection training framework to obtain a target detection set;
and the riding detection module (4) is used for detecting the target detection set by utilizing a riding detection algorithm to obtain a target detection result.
9. A helmet detection electronic device based on improved YOLOv5s, comprising: the system comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus;
a memory for storing a computer program;
a processor for implementing the method steps of any one of claims 1-7 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the improved YOLOv5 s-based helmet detection method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110144067.2A CN112966810B (en) | 2021-02-02 | 2021-02-02 | Helmet detection method and device based on improved YOLOv5s, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110144067.2A CN112966810B (en) | 2021-02-02 | 2021-02-02 | Helmet detection method and device based on improved YOLOv5s, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112966810A true CN112966810A (en) | 2021-06-15 |
| CN112966810B CN112966810B (en) | 2023-07-11 |
Family
ID=76271920
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110144067.2A Active CN112966810B (en) | 2021-02-02 | 2021-02-02 | Helmet detection method and device based on improved YOLOv5s, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112966810B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114120366A (en) * | 2021-11-29 | 2022-03-01 | 上海应用技术大学 | Non-motor vehicle helmet detection method based on generation countermeasure network and yolov5 |
| CN114170161A (en) * | 2021-11-25 | 2022-03-11 | 云南大学 | Cigarette appearance defect detection method and system |
| CN114418064A (en) * | 2021-12-27 | 2022-04-29 | 西安天和防务技术股份有限公司 | Target detection method, terminal equipment and storage medium |
| CN114943697A (en) * | 2022-05-16 | 2022-08-26 | 福建农林大学 | Tomato maturity detection method based on lightweight improved YOLO |
| CN115205825A (en) * | 2022-07-15 | 2022-10-18 | 宜宾学院 | Improved YOLOV 5-based traffic sign detection and identification method for driving video sequence images |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190392323A1 (en) * | 2018-06-22 | 2019-12-26 | Moffett AI, Inc. | Neural network acceleration and embedding compression systems and methods with activation sparsification |
| US20200005468A1 (en) * | 2019-09-09 | 2020-01-02 | Intel Corporation | Method and system of event-driven object segmentation for image processing |
| CN110991362A (en) * | 2019-12-06 | 2020-04-10 | 西安电子科技大学 | Pedestrian detection model based on attention mechanism |
| CN111209921A (en) * | 2020-01-07 | 2020-05-29 | 南京邮电大学 | License plate detection model based on improved YOLOv3 network and construction method |
| AU2020100705A4 (en) * | 2020-05-05 | 2020-06-18 | Chang, Jiaying Miss | A helmet detection method with lightweight backbone based on yolov3 network |
| CN111814762A (en) * | 2020-08-24 | 2020-10-23 | 深延科技(北京)有限公司 | Helmet wearing detection method and device |
-
2021
- 2021-02-02 CN CN202110144067.2A patent/CN112966810B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190392323A1 (en) * | 2018-06-22 | 2019-12-26 | Moffett AI, Inc. | Neural network acceleration and embedding compression systems and methods with activation sparsification |
| US20200005468A1 (en) * | 2019-09-09 | 2020-01-02 | Intel Corporation | Method and system of event-driven object segmentation for image processing |
| CN110991362A (en) * | 2019-12-06 | 2020-04-10 | 西安电子科技大学 | Pedestrian detection model based on attention mechanism |
| CN111209921A (en) * | 2020-01-07 | 2020-05-29 | 南京邮电大学 | License plate detection model based on improved YOLOv3 network and construction method |
| AU2020100705A4 (en) * | 2020-05-05 | 2020-06-18 | Chang, Jiaying Miss | A helmet detection method with lightweight backbone based on yolov3 network |
| CN111814762A (en) * | 2020-08-24 | 2020-10-23 | 深延科技(北京)有限公司 | Helmet wearing detection method and device |
Non-Patent Citations (3)
| Title |
|---|
| YE LI 等: "A Multi-task Joint Framework for Real-time Person Search" * |
| 舒军 等: "基于改进YOLOv3的航拍目标实时检测方法" * |
| 解怀奇 等: "基于通道注意力机制的视频人体行为识别" * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114170161A (en) * | 2021-11-25 | 2022-03-11 | 云南大学 | Cigarette appearance defect detection method and system |
| CN114120366A (en) * | 2021-11-29 | 2022-03-01 | 上海应用技术大学 | Non-motor vehicle helmet detection method based on generation countermeasure network and yolov5 |
| CN114120366B (en) * | 2021-11-29 | 2023-08-25 | 上海应用技术大学 | Non-motor helmet detection method based on generation of countermeasure network and yolov5 |
| CN114418064A (en) * | 2021-12-27 | 2022-04-29 | 西安天和防务技术股份有限公司 | Target detection method, terminal equipment and storage medium |
| CN114943697A (en) * | 2022-05-16 | 2022-08-26 | 福建农林大学 | Tomato maturity detection method based on lightweight improved YOLO |
| CN115205825A (en) * | 2022-07-15 | 2022-10-18 | 宜宾学院 | Improved YOLOV 5-based traffic sign detection and identification method for driving video sequence images |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112966810B (en) | 2023-07-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112966810A (en) | Helmet detection method and device based on improved YOLOv5s, electronic equipment and storage medium | |
| CN108053427B (en) | An improved multi-target tracking method, system and device based on KCF and Kalman | |
| CN109460699B (en) | Driver safety belt wearing identification method based on deep learning | |
| CN106778583B (en) | Vehicle attribute recognition method and device based on convolutional neural network | |
| CN112084866A (en) | Target detection method based on improved YOLO v4 algorithm | |
| CN111832461B (en) | Method for detecting wearing of non-motor vehicle riding personnel helmet based on video stream | |
| CN106354816A (en) | Video image processing method and video image processing device | |
| CN111242015A (en) | A Method for Predicting Driving Dangerous Scenes Based on Motion Contour Semantic Map | |
| CN107292933B (en) | Vehicle color identification method based on BP neural network | |
| CN105760847B (en) | A kind of visible detection method of pair of helmet of motorcycle driver wear condition | |
| CN117333753B (en) | Fire detection method based on PD-YOLO | |
| CN114298187B (en) | An Object Detection Method Fused with Improved Attention Mechanism | |
| CN116071374B (en) | A lane line instance segmentation method and system | |
| CN110689578A (en) | An obstacle recognition method for UAV based on monocular vision | |
| CN111428735A (en) | Truck brand classification method based on migration learning deep network fusion model | |
| CN112906533B (en) | Safety helmet wearing detection method based on self-adaptive detection area | |
| CN118918541A (en) | Electric vehicle helmet wearing detection method based on improvement YOLOv s | |
| CN119229374A (en) | A lightweight and dense pedestrian detection method and system based on YOLOv8 | |
| Montenegro et al. | Pedestrian detection at daytime and nighttime conditions based on YOLO-v5 | |
| CN112907972A (en) | Road vehicle flow detection method and system based on unmanned aerial vehicle and computer readable storage medium | |
| Wang et al. | Prb-fpn+: Video analytics for enforcing motorcycle helmet laws | |
| CN115205813A (en) | Multi-view road vehicle matching system and method based on convolutional neural network | |
| CN108921147B (en) | Black smoke vehicle identification method based on dynamic texture and transform domain space-time characteristics | |
| Sun et al. | Flame Image Detection Algorithm Based onComputer Vision. | |
| CN119600670A (en) | A face mask detection method based on YOLOv8 to resist occlusion and counterfeit attacks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |