CN115984784A - Safety incident identification system and method for dangerous approach of ramp vehicles for airport supervision - Google Patents

Abstract

The invention relates to a safety event identification system and method for the dangerous approach of apron vehicles used for airport supervision, which belongs to the field of airport safety prevention; it includes a video stream access module for real-time access and processing of camera video data; an interface processing module for Access production and safety system data; algorithm processing module, used to identify aircraft and vehicle data in the real-time video stream of the camera; front-end push module, used to push the processing results of safety events to the front-end application for display. The advantages of the present invention are: a large number of cameras are installed at the airport, and the video stream information of the cameras is collected in real time. By using the deep learning convolution neural network algorithm technology, information such as vehicles and aircraft in the video streams of the cameras is identified, and the distance of the identification target information is calculated to determine whether there is Danger is close to a safety risk. The advantages of the present invention: the present invention will significantly improve the problem of safety risk omissions existing in manual patrol inspection, and significantly improve the safety management level of the airport.

Description

Airport monitoring airport vehicle danger approaching safety event recognition system and method thereof

Technical Field

The invention relates to a airport supervision apron vehicle danger approach safety event recognition system and a recognition method thereof based on video analysis, belonging to the field of airport safety precaution.

Background

The dangerous security incident that is close as airport safety risk level is higher of vehicle, mainly rely on the manual work to patrol and examine at present most airport, inefficiency just has obvious safety risk missing item, some airports install GPS location facility for the vehicle, rely on vehicle positional information to carry out the dangerous judgement foundation that is close, but GPS mobile unit cost is higher, the installation scope is limited, the vehicle longitude and latitude location exists the drift, the trip point phenomenon, there is certain identification error, in practical application, the judgement foundation that is close as the dangerous vehicle is comprehensive inadequately, misjudgement problem can appear often, the violation punishment foundation of complete identification is difficult to carry out.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides a airport-supervision-based airport vehicle danger approaching safety event identification system and a method thereof, and the technical scheme of the invention is as follows:

a airport supervision apron vehicle dangerous approaching safety event recognition system comprises a video stream access module, a video data real-time access processing module and a safety monitoring module, wherein the video stream access module is used for accessing and processing video data of a camera in real time;

the interface processing module is used for accessing production system data and safety system data;

the algorithm processing module is used for identifying aircraft and vehicle data in the camera real-time video stream, comprehensively judging the identified target data and identifying possible vehicle danger approaching safety events;

and the front-end pushing module is used for pushing the processing result of the security event to the front-end application for display.

The algorithm processing module uses a deep learning convolutional neural network algorithm, and the structure is as follows: input → [ conv → relu ]. N → [ pool ]. M → [ fc → relu ]. K;

wherein input represents an input layer, an image is input into a convolutional neural network after being processed, conv represents a convolutional layer and is used for extracting features in the image, relu represents an activation function of the convolutional layer, n represents that n convolutional layers exist, pool represents a pooling layer and is used for reducing dimensions of a feature map, the number of parameters is compressed, the capability of model generalization is improved, overfitting is prevented, m represents that m pooling layers exist, fc represents a fully-connected layer, the features of the convolutional layer, the activation layer and the pooling layer are integrated, relu represents an activation function of the fully-connected layer, and k represents that k fully-connected layers exist;

the convolutional neural network algorithm core is a Loss function, corresponding forward propagation and backward propagation are carried out through the Loss function, a corresponding convolutional neural network is trained, and a real scene is matched.

After identifying specific aircraft and vehicle target information, the algorithm processing module stores the identified aircraft and vehicle pixel point sets in a list to be calculated, traverses and compares every two targets in a traversing mode, takes out each target point set, calculates a geometric center, and assumes that a current point set is a rectangular point set and a target coordinate is: { (x 1, y 1), (x 2, y)2) (x 3, y 3), (x 4, y 4) }, the coordinate p of the center point of the target object is:

suppose that the coordinates of four vertex pixels arranged clockwise in the rectangular area identified by the camera are as follows: { (mx 1, my 1), (mx 2, my 2), (mx 3, my 3), (mx 4, my 4) }, the corresponding longitude and latitude coordinates are: { (pmx 1, pmy 1), (pmx 2, pmy 2), (pmx 3, pmy 3), (pmx 4, pmy 4) }, the X-axis transformation parameter XC is:

the Y-axis conversion parameter YC is:

assuming that the pixel coordinates of the central points of the two comparison targets are (px 1, py 1), (px 2, py 2), respectively, the true distance L between the central points of the two targets is:

early warning is carried out by judging whether the distance information is lower than the set distance of the system, and if the two compared target objects are an aircraft and an aircraft respectively, the distance parameter is PC; if the two comparison target objects are respectively an aircraft and a vehicle, the distance parameter is PCC; if the two comparison target objects are respectively a vehicle and a vehicle, the distance parameter is CC; the real judgment distance is related to the relative positions of the two comparative target objects, and the distances are respectively judged to be

And &>

In the actual judgment, whether the real distance between the central points of the two target objects is smaller than the judgment distance is compared in real time, if so, a safety risk event is generated, and relevant early warning is carried out.

A method for recognizing dangerous approaching safety events of vehicles on airport-based apron for supervision comprises the following steps:

s1, accessing and processing video data of a camera in real time;

s2, accessing production system data and safety system data;

s3, recognizing aircraft and vehicle data in the real-time video stream of the camera, comprehensively recognizing the recognized target data, judging, and recognizing possible vehicle danger approaching safety events;

and S4, pushing the processing result of the security event to a front-end application for display.

The invention has the advantages that: the method comprises the steps of installing a large number of cameras in an airport, collecting camera video stream information in real time, identifying information such as vehicles and aircrafts in the camera video stream by adopting a deep learning convolutional neural network algorithm technology, and judging whether dangerous approaching safety risks exist or not by calculating identification target information distance. The method can obviously improve the problem of missing safety risks in manual inspection, and can be used as a powerful supplement for a GPS vehicle-mounted terminal judgment method to obviously improve the airport safety management level.

Drawings

FIG. 1 is a block diagram of a airport supervisory apron vehicle hazardous proximity safety event identification system of the present invention.

Detailed Description

The invention is further described below in conjunction with specific embodiments, and the advantages and features of the invention will become more apparent as the description proceeds. These examples are illustrative only and do not limit the scope of the present invention in any way. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention, and that such changes and modifications may be made without departing from the spirit and scope of the invention.

Referring to fig. 1, the invention relates to a airport supervision apron vehicle danger approach safety event recognition system, which comprises a video stream access module 1, wherein a camera on an airport apron real-time code stream data is accessed to the module for video transcoding processing, and a related video is pushed to an algorithm processing module for recognition;

the interface processing module 2 is accessed to different association systems of airport production and security, so as to better provide field condition information for the algorithm processing module to carry out more comprehensive logic judgment;

the algorithm processing module 3 is used for identifying aircraft and vehicle data in the camera real-time video stream, comprehensively judging the identified target data and identifying possible vehicle danger approaching safety events;

and the front-end pushing module 4 is used for pushing the processing result of the security event to the front-end application for display.

The algorithm processing module 3 uses a deep learning convolutional neural network algorithm, input → [ conv → relu ] → [ pool ] → [ m ] → [ fc → relu ]. K, input represents an input layer, the input layer inputs the convolutional neural network after processing the image, conv represents a convolutional layer and is used for extracting the features in the image, relu represents an activation function of the convolutional layer, n represents the existence of n convolutional layers, pool represents a pooling layer and is used for reducing the dimension of the feature map, the number of compression parameters improves the capability of model generalization and prevents overfitting, m represents the existence of m pooling layers, fc represents a fully-connected layer, and the features extracted by the convolutional layers, the activation layer and the pooling layer are integrated, relu represents an activation function of the fully-connected layer, and k represents the existence of a k fully-connected layer.

After the algorithm processing module 3 identifies specific aircraft and vehicle target information, the identified aircraft and vehicle pixel point sets are stored in a list to be calculated, two targets are subjected to traversal comparison, the geometric center of each target is calculated, and the target coordinates are assumed as: { (x 1, y 1), (x 2, y 2), (x 3, y 3), (x 4, y 4) }, the target center point coordinate p is:

suppose that the coordinates of four vertex pixels arranged clockwise in the rectangular area recognized by the camera are as follows: { (mx 1, my 1), (mx 2, my 2), (mx 3, my 3), (mx 4, my 4) }, the corresponding longitude and latitude coordinates are: { (pmx 1, pmy 1), (pmx 2, pmy 2), (pmx 3, pmy 3), (pmx 4, pmy 4) }, the X-axis transformation parameter XC is:

the Y-axis conversion parameter YC is:

Assuming that the pixel coordinates of the central points of the two comparison targets are (px 1, py 1), (px 2, py 2), respectively, the true distance L between the central points of the two targets is:

early warning is carried out by judging whether the distance information is lower than the set distance of the system, if the two comparison target objects are respectively an aircraft and an aircraft, the distance parameter is PC, if the two comparison target objects are respectively the aircraft and the vehicle, the distance parameter is PCC, if the two comparison target objects are respectively the vehicle and the vehicle, the distance parameter is CC, the distance is really judged to be related to the relative positions of the two comparison target objects, and the distance is judged to be PC

In the actual judgment, whether the real distance between the central points of the two target objects is smaller than the judgment distance is compared in real time, if so, a safety risk event is generated, and relevant early warning is carried out.

The invention also relates to an identification method based on the airport supervision apron vehicle danger approach safety event identification system, which comprises the following steps:

s1, accessing and processing video data of a camera in real time;

s2, accessing production system data and safety system data;

s3, recognizing aircraft and vehicle data in the real-time video stream of the camera, comprehensively recognizing the recognized target data for judgment, and recognizing possible vehicle danger approaching safety events;

and S4, pushing the processing result of the security event to a front-end application for display.

The step S3 is specifically: target identification using deep learning convolutional neural network algorithm, input → [ conv → relu]*n→[pool]*m→[fc→relu]* k. The identified target objects are traversed and compared with each other, and the central coordinates are solved into

Suppose that the coordinates of four vertex pixels arranged clockwise in the rectangular area recognized by the camera are as follows: { (mx 1, my 1), (mx 2, my 2), (mx 3, my 3), (mx 4, my 4) }, the corresponding longitude and latitude coordinates are: { (pmx 1, pmy 1), (pmx 2, pmy 2), (pmx 3, pmy 3), (pmx 4, pmy 4) }, the X-axis transformation parameter XC is:

the Y-axis conversion parameter YC is:

Then the real distance L between the center points of the two targets is:

and comparing whether the real distance between the central points of the two target objects is smaller than the judgment distance in real time, and if so, generating a safety risk event and carrying out related early warning.

The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered as the technical solutions and the inventive concepts of the present invention within the technical scope of the present invention.

Claims (4)

1. A airport vehicle dangerous approaching safety event identification system for airport supervision is characterized by comprising

The video stream access module is used for accessing and processing the video data of the camera in real time;

the interface processing module is used for accessing production system data and safety system data;

the algorithm processing module is used for identifying aircraft and vehicle data in the camera real-time video stream, comprehensively judging the identified target data and identifying possible vehicle danger approaching safety events;

and the front-end pushing module is used for pushing the processing result of the security event to the front-end application for display.

2. The airport supervisory apron vehicle hazard proximity safety event identification system of claim 1, wherein said algorithm processing module uses a deep learning convolutional neural network algorithm configured as follows: input → [ conv → relu ]. N → [ pool ]. M → [ fc → relu ]. K;

wherein input represents an input layer, an image is input into a convolutional neural network after being processed, conv represents a convolutional layer and is used for extracting features in the image, relu represents an activation function of the convolutional layer, pool represents a pooling layer and is used for reducing the dimension of the feature map, compressing the number of parameters, improving the capability of model generalization and preventing overfitting, fc represents a full-connection layer, the features extracted by the convolutional layer, the activation layer and the pooling layer are integrated, and relu represents the activation function of the full-connection layer;

the convolutional neural network algorithm core is a Loss function, corresponding forward propagation and backward propagation are carried out through the Loss function, a corresponding convolutional neural network is trained, and a real scene is matched.

3. The airport supervisory apron vehicle danger approach safety event identification system of claim 2, wherein the algorithm processing module stores the identified aircraft and vehicle pixel point sets in a list to be calculated after identifying specific aircraft and vehicle target information, performs traversal comparison on two targets by adopting a traversal mode, takes out each target point set, calculates a geometric center, and assumes that the target coordinates are: { (x 1, y 1), (x 2, y 2), (x 3, y 3), (x 4, y 4) }, the target center point coordinate p is:

suppose that the coordinates of four vertex pixels arranged clockwise in the rectangular area identified by the camera are as follows: { (mx 1, my 1), (mx 2, my 2), (mx 3, my 3), (mx 4, my 4) }, corresponding latitude and longitude coordinates are: { (pmx 1, pmy 1), (pmx 2, pmy 2), (pmx 3, pmy 3), (pmx 4, pmy 4) }, the X-axis transformation parameter XC is:

the Y-axis conversion parameter YC is:

Assuming that the pixel coordinates of the central points of the two comparison targets are (px 1, py 1), (px 2, py 2), respectively, the true distance L between the central points of the two targets is:

early warning is carried out by judging whether the distance information is lower than the set distance of the system, and if the two comparison target objects are respectively an aircraft and an aircraft, the distance parameter is PC; if the two comparison target objects are respectively an aircraft and a vehicle, the distance parameter is PCC; if the two comparison target objects are respectively a vehicle and a vehicle, the distance parameter is CC; the real judgment distance is related to the relative positions of the two comparative target objects, and the distances are respectively judged to be

And &>

In the actual judgment, whether the real distance between the central points of the two target objects is smaller than the judgment distance is compared in real time, if so, a safety risk event is generated, and relevant early warning is carried out.

4. A method for identifying a danger proximity safety incident for an apron vehicle based on any one of claims 1 to 3, characterized in that it comprises the following steps:

s1, accessing and processing video data of a camera in real time;

s2, accessing production system data and safety system data;

s3, recognizing aircraft and vehicle data in the real-time video stream of the camera, comprehensively recognizing the recognized target data for judgment, and recognizing possible vehicle danger approaching safety events;

and S4, pushing the processing result of the security event to a front-end application for display.

Images