CN115773720A - A device and method for measuring underwater target size based on optical vision - Google Patents

Abstract

The invention relates to a device and method for measuring the size of an underwater target based on optical vision. The device includes an underwater camera, a plurality of underwater lasers, a turntable and a data processing module. The method includes: establishing a relationship between a spot image and the underwater camera The mapping relationship between the distance D between them and the perimeter L of the spot pattern; the turntable is transferred from the starting position D ₁ of the underwater target to the end position D ₂ of the underwater target, and the rotation angle α of the turntable is recorded, and at the same time Take the spot images on the surface of the underwater target at _D1 and _D2 ; use the data processing module to process the spot images corresponding to _D1 and _D2 to obtain the perimeter _L1 and _L2 of the spot images; pass between D and L The mapping relationship of D ₁ and D ₂ is obtained; the actual size M of the underwater target is obtained by calculation. The invention can be used for measuring the size of an underwater target, has the characteristics of simple operation, good adaptability to the water quality environment, high precision, simple structure and low complexity, and is convenient for popularization and application.

Description

Device and method for measuring underwater target size based on optical vision

technical field

The invention relates to the field of underwater optical measurement, in particular to an optical vision-based underwater target size measurement device and method.

Background technique

The size measurement of underwater targets is very important for applications such as marine pipeline inspection, deep sea engineering operations, and lake and dam detection. Due to the serious attenuation of radio waves in water, traditional electromagnetic detection methods cannot adapt to the underwater environment. At present, the commonly used underwater target size measurement mainly relies on acoustic methods and optical methods. Among them, the acoustic measurement is to measure the size of the target in the water through the sound wave, which has the advantage of a long working distance; however, due to the limitation of the large wavelength of the sound wave, this method has limitations in the measurement accuracy. In contrast, the optical method has the advantages of high precision and has attracted extensive attention in recent years.

Optics-based underwater target size measurement methods mainly include Time Of Flight (TOF) measurement, underwater binoculars, underwater laser scanning imaging, underwater active laser dot matrix projection imaging, etc. The TOF method is to obtain distance information by measuring the time difference between laser emission and echo detection, and then form a point cloud by scanning to measure the size of the target; however, this method is easily affected by seawater backscattering, thus limiting the application of this technology scope.

Underwater binocular measurement is to establish the three-dimensional structure of the target through the method of stereo vision, and then obtain the size information of the target. Please refer to "Zhang Honglong, Chen Tao, Zhuang Peiqin, etc., "Research on Underwater Three-dimensional Measurement System Based on Stereo Vision", here The disadvantage of this method is that the measurement effect is greatly affected by the turbidity and illuminance of the water body, and the measurement distance and stability are limited.

Underwater laser scanning imaging is to capture the influence of the line laser on the surface of the target through the camera, and then obtain the three-dimensional information of the target surface through the method of triangulation; Refer to "Xie Xiaomeng, "Analysis of Underwater Target Laser Scanning Depth Detection Accuracy".

The underwater laser dot matrix projection imaging measurement system is mainly composed of array lasers and cameras parallel to each other. When working, the array laser projects a dot matrix composed of laser dots into the water. The camera extracts the dot matrix features, and then obtains the distance and distance of the target object. For size information, see "Yang Mengning, Han Biao, Zhang Bing, etc., "An Underwater Measurement Equipment and Underwater Measurement Method""; this method is helpful to realize the size measurement of long-distance targets, but the target Size is limited by the camera's field of view.

Contents of the invention

In view of the above-mentioned problems in the prior art, the first technical problem to be solved by the present invention is to provide a device for measuring underwater targets.

The second technical problem to be solved is to provide a measurement method with high measurement accuracy and wide application range for underwater targets.

In order to solve the above-mentioned first technical problem, the present invention adopts the following technical solution: a device for measuring the size of an underwater target based on optical vision, including an underwater camera, J underwater lasers, a turntable and a data processing module, wherein J≥ 2;

The J underwater lasers are used to form a spot array on the surface of the underwater target, and the J beams emitted by the J underwater lasers are parallel to each other, and the beams of the underwater lasers are directed to the optical axis of the underwater camera parallel;

The underwater camera is used to photograph the spot pattern formed by the light beam on the surface of the underwater target;

The turntable is used to support the underwater camera and J underwater lasers;

The data processing module is used to identify the light spot pattern captured by the underwater camera, and obtain the distance information between the underwater camera and the underwater target through the imaging principle of "near large, far small", and combine the rotation of the turntable angle to calculate the size information of the underwater target.

Preferably, the data processing module includes an area segmentation module and a measurement calculation module, and the area segmentation module and the measurement calculation module are sequentially connected;

The area segmentation module is a Mask R-CNN segmentation algorithm model, and the measurement calculation module uses the area information obtained by the area segmentation module to calculate the size of the underwater target.

In order to solve the above-mentioned second technical problem, the present invention adopts the following technical solution: a method for measuring the size of an underwater target based on optical vision, characterized in that: the underwater target size measurement based on optical vision described in claim 1 is adopted The device, described measuring method comprises the steps:

S1: Let D be the distance between the spot image and the underwater camera, L be the circumference of the spot pattern, and establish the mapping relationship between D and L, the expression is as follows:

D=a×L ^b +c; (1)

Among them, a, b and c represent constants;

S2: The turntable turns from the initial position _D1 of the underwater target to the end position _D2 of the underwater target, and records the rotation angle α of the turntable; at the same time, the underwater camera shoots the underwater targets at _D1 and _D2 Spot image on the surface of the object;

S3: use the data processing module to process the spot image corresponding to _D1 to obtain the perimeter _L1 of the spot image; use the data processing module to process the spot image corresponding to _D2 to obtain the perimeter _L2 of the spot image;

S4: Substituting L ₁ and L ₂ into formula (1) for calculation to obtain D ₁ and D ₂ respectively;

S5: Calculate the actual size M of the underwater target, the specific expression is as follows:

Among them, α represents the angle of rotation of the turntable.

As a preference, the specific steps for establishing the mapping relationship between D and L in the S1 are as follows:

S11: Acquire W spot images with distance labels;

S12: Using the spot image as the input of the data processing module, obtain the pixel coordinates of multiple spots, and use the pixel coordinates to calculate the perimeter L of the spot image, and the specific expression of L is as follows:

Among them, (x _r , y _r ) represents the pixel coordinates of the light spot, and r represents the number of light spots;

S13: Form a data set with perimeter L of all spot images with distance labels D, and D and L correspond one-to-one;

S14: Fitting D and L in the data set by the least square method to obtain a functional mapping relationship between D and L.

As a preference, the pixel coordinates of multiple light spots are acquired in the S12, and the specific steps of calculating the perimeter L of the light spot image are as follows:

S121: using the W spot images acquired in S11 as a training set;

S122: Carry out geometric figure annotation on the W spot images respectively, and obtain W light spot images with graphic annotation labels, and each light spot image with graphic annotation labels is a training sample;

S123: let t=1;

S124: Select the tth training sample from W as the input of the Mask R-CNN neural network model, and output the Mask mask segmentation image t' corresponding to the tth training sample;

S125: Perform contour fitting and edge detection processing on t' to obtain the predicted image t" of the tth training sample;

S126: Set the loss threshold, and calculate the loss Loss between t" and t;

When the Loss value is less than the loss threshold, the trained Mask R-CNN neural network model is obtained, and the next step is performed; otherwise, the parameters in the Mask R-CNN neural network model are updated by backpropagation, and t=t+1 is set, And return to S124;

S127: Marking the image of the to-be-predicted light spot with a geometric figure, and using it as an input of the trained Mask R-CNN neural network model, and outputting the pixel coordinates of the light spot surrounding the image of the to-be-predicted light spot;

S128: Calculate the perimeter L of the to-be-predicted spot image by using the spot pixel coordinates obtained in S127.

Compared with the prior art, the present invention has at least the following advantages:

1. Use the existing Mask R-CNN neural network to accurately divide the spot area on the underwater target, so that the range of the area can be clarified, and there are accurate pixel coordinates; then the perimeter of the spot image can be obtained through the pixel coordinates Long, using the mapping relationship between the distance D between the underwater camera and the underwater target and the perimeter L of the spot image, the exact starting position D ₁ and the ending position D ₂ of the beam point can be obtained, and finally using D ₁ and D ₂ to calculate the size of the underwater target, such a measurement process is more adaptable to different water quality environments, and also makes the measurement results more accurate.

2. Use the turntable to carry the camera and laser so that it can rotate, combined with the method of triangulation, the monitoring range of the target size can be improved, and it is more suitable for the measurement of large underwater targets.

3. Using deep learning to detect the position information of the light spot can effectively suppress the influence of backscattering of the water body on the light spot detection, and help to improve the detection accuracy of the light spot position.

Description of drawings

Fig. 1 is a schematic diagram of the underwater measuring device of the present invention.

Fig. 2 is a schematic diagram of the underwater measurement method of the present invention.

Fig. 3 is a schematic diagram of the measurement system in the embodiment of the present invention.

Fig. 4 is a schematic diagram of the underwater distance measuring device adopted in the embodiment of the present invention.

Fig. 5 is a testing device composed of an underwater camera and four underwater lasers in the embodiment of the present invention.

Fig. 6 Recognition of underwater light spots in the embodiment of the present invention.

Fig. 7 is the distance of test point A in the embodiment of the present invention.

Fig. 8 is the distance of test point B in the embodiment of the present invention.

Detailed ways

The present invention will be described in further detail below.

Referring to Figures 1-2, a device for measuring the size of an underwater target based on optical vision, including an underwater camera, J underwater lasers, a turntable and a data processing module, wherein J≥2;

The J underwater lasers are used to form a spot array on the surface of the underwater target, and the J beams emitted by the J underwater lasers are parallel to each other, and the beams of the underwater lasers are directed to the optical axis of the underwater camera parallel;

The underwater camera is used to photograph the spot pattern formed by the light beam on the surface of the underwater target;

The turntable is used to support the underwater camera and J underwater lasers, the purpose of which is to enable the J underwater lasers and underwater cameras to rotate as a whole;

The data processing module is used to identify the light spot pattern captured by the underwater camera, and obtain the distance information between the underwater camera and the underwater target through the imaging principle of "near large, far small", and combine the rotation of the turntable angle to calculate the size information of the underwater target.

The data processing module includes an area segmentation module and a measurement calculation module, the area segmentation module and the measurement calculation module are sequentially connected; the data processing module is used to analyze and calculate the laser spot array image captured by the underwater camera, The main thing is to obtain the distance information between the camera and the target through the imaging principle of "near large, far small", and combine the rotation angle of the turntable to calculate the size information of the target;

The area segmentation module is a Mask R-CNN segmentation algorithm model, and the Mask R-CNN segmentation algorithm model is the prior art, and the measurement and calculation module uses the area information obtained by the area segmentation module to calculate the size of the underwater target.

For specific implementation, see Fig. 3, the measurement system for the embodiment of this project mainly consists of four laser arrays, cameras, turntables, computers, test point A and test point B. Among them, four lasers and cameras are placed in a watertight housing, as shown in Figure 4; the laser emission direction and the optical axis of the camera are parallel to each other, and when the laser is irradiated on the surface of the target, the laser spot can be observed from the camera. The turntable is placed in the water and can rotate at a certain angle according to the command, which is used to make the device composed of the laser array and the camera rotate in the water as a whole. The equipment composed of computer, turntable, laser and camera is connected by watertight cable, which is used to read the rotation angle of the turntable and the laser light spot pictures captured by the camera, and calculate the size information of the target. Test point A and test point B are used to simulate the start position and end position of the object to be tested respectively.

As shown in Figure 5, the underwater camera, underwater laser and the corresponding mounting bracket are put into the water for calibration, and the spot images at 1.5m to 11m are collected at a distance of 0.05m; these tubesheet images form a data set;

S1: Let D be the distance between the spot image and the underwater camera, L be the circumference of the spot pattern, and establish the mapping relationship between D and L, the expression is as follows:

D=a×L ^b +c; (1)

Among them, a, b and c represent constants;

The specific steps for establishing the mapping relationship between D and L in the S1 are as follows:

S11: Acquire W spot images with distance labels;

S12: Using the spot image as the input of the data processing module, obtain the pixel coordinates of multiple spots, and use the pixel coordinates to calculate the perimeter L of the spot image, and the specific expression of L is as follows:

Among them, (x _r , y _r ) represents the pixel coordinates of the light spot, and r represents the number of light spots;

S121: using the W spot images acquired in S11 as a training set;

S122: Carry out geometric figure annotation on the W spot images respectively, and obtain W light spot images with graphic annotation labels, and each light spot image with graphic annotation labels is a training sample;

S123: let t=1;

S124: Select the tth training sample from W as the input of the Mask R-CNN neural network model, and output the Mask mask segmentation image t' corresponding to the tth training sample;

S125: Perform contour fitting and edge detection processing on t', both contour fitting and edge detection are existing technologies, and obtain the predicted image t" of the tth training sample;

S126: Set the loss threshold, and calculate the loss Loss between t" and t;

When the Loss value is less than the loss threshold, the trained Mask R-CNN neural network model is obtained, and the next step is performed; otherwise, the parameters in the Mask R-CNN neural network model are updated by backpropagation, and t=t+1 is set, And return to S124;

S127: Marking the image of the to-be-predicted light spot with a geometric figure, and using it as an input of the trained Mask R-CNN neural network model, and outputting the pixel coordinates of the light spot surrounding the image of the to-be-predicted light spot;

S128: Calculate the perimeter L of the to-be-predicted spot image by using the spot pixel coordinates obtained in S127.

S13: Form a data set with perimeter L of all spot images with distance labels D, and D and L correspond one-to-one;

S14: Fitting D and L in the data set by a least square method to obtain a functional mapping relationship between D and L, the least square method is a prior art;

S2: The turntable turns from the initial position _D1 of the underwater target to the end position _D2 of the underwater target, and records the rotation angle α of the turntable; at the same time, the underwater camera shoots the underwater targets at _D1 and _D2 Spot image on the surface of the object;

S3: use the data processing module to process the spot image corresponding to _D1 to obtain the perimeter _L1 of the spot image; use the data processing module to process the spot image corresponding to _D2 to obtain the perimeter _L2 of the spot image;

S4: Substituting L ₁ and L ₂ into formula (1) for calculation to obtain D ₁ and D ₂ respectively;

S5: Calculate the actual size M of the underwater target, the specific expression is as follows:

Among them, α represents the angle of rotation of the turntable.

In the specific implementation, as shown in Figure 6, the weight of the model is trained through the Mask R-CNN deep learning algorithm, which is used to accurately identify and segment the spot area, and fit the relationship between the perimeter and distance of the image composed of the center of the spot in the camera , such as formula (4). Where L is the perimeter of the measured spot area, in pixel; D is the predicted distance between the target plane and the camera, in cm

D＝99556.57L ^-0.9753 (4)

Finally, as shown in Figure 3, set up a test system in water, use test point A and test point B to simulate the start position and end position of the target to be tested respectively, and the distance between the two is about 490cm, that is, test point A and test point B The simulated target size is approximately 490 cm. The angle at which the turntable rotates from test point A to test point B is 45°. As shown in Figure 7, the distance between the camera and test point A is 168.91cm, and the distance between test point B is 591.54cm. The calculated simulated target size between test point A and test point B is 486.98cm. The relative error of target measurement is about 0.62%. This verifies the feasibility of the method.

Finally, it is noted that the above embodiments are only used to illustrate the technical solutions of the present invention without limitation. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be carried out Modifications or equivalent replacements without departing from the spirit and scope of the technical solution of the present invention shall be covered by the claims of the present invention.

Claims (5)

1. A device for measuring the size of an underwater target based on optical vision, characterized in that it includes an underwater camera, J underwater lasers, a turntable and a data processing module, wherein J≥2; The J underwater lasers are used to form a spot array on the surface of the underwater target, and the J beams emitted by the J underwater lasers are parallel to each other, and the beams of the underwater lasers are directed to the optical axis of the underwater camera parallel; The underwater camera is used to photograph the spot pattern formed by the light beam on the surface of the underwater target; The turntable is used to support the underwater camera and J underwater lasers; The data processing module is used to identify the light spot pattern captured by the underwater camera, and obtain the distance information between the underwater camera and the underwater target through the imaging principle of "near large, far small", and combine the rotation of the turntable angle to calculate the size information of the underwater target. 2. a kind of device based on the underwater target size measurement of optical vision as claimed in claim 1, is characterized in that: described data processing module comprises area segmentation module and measurement calculation module, and described area segmentation module and measurement calculation module for sequential connection; The area segmentation module is a Mask R-CNN segmentation algorithm model, and the measurement calculation module uses the area information obtained by the area segmentation module to calculate the size of the underwater target. 3. A method for measuring the size of an underwater target based on optical vision, characterized in that: the device for measuring the size of an underwater target based on optical vision as claimed in claim 1 is used, and the measuring method comprises the steps: S1: Let D be the distance between the spot image and the underwater camera, L be the circumference of the spot pattern, and establish the mapping relationship between D and L, the expression is as follows: D=a×L ^b +c; (1) Among them, a, b and c represent constants; S2: The turntable turns from the initial position _D1 of the underwater target to the end position _D2 of the underwater target, and records the rotation angle α of the turntable; at the same time, the underwater camera shoots the underwater targets at _D1 and _D2 Spot image on the surface of the object; S3: use the data processing module to process the spot image corresponding to _D1 to obtain the perimeter _L1 of the spot image; use the data processing module to process the spot image corresponding to _D2 to obtain the perimeter _L2 of the spot image; S4: Substituting L ₁ and L ₂ into formula (1) for calculation to obtain D ₁ and D ₂ respectively; S5: Calculate the actual size M of the underwater target, the specific expression is as follows:

Among them, α represents the angle of rotation of the turntable. 4. the method for a kind of underwater target size measurement based on optical vision as claimed in claim 3 is characterized in that: the specific steps of establishing the mapping relationship between D and L among the described S1 are as follows: S11: Acquire W spot images with distance labels; S12: Using the spot image as the input of the data processing module, obtain the pixel coordinates of multiple spots, and use the pixel coordinates to calculate the perimeter L of the spot image, and the specific expression of L is as follows:

Among them, (x _r , y _r ) represents the pixel coordinates of the light spot, and r represents the number of light spots; S13: Form a data set with perimeter L of all spot images with distance labels D, and D and L correspond one-to-one; S14: Fitting D and L in the data set by the least square method to obtain a functional mapping relationship between D and L. 5. the method for a kind of underwater target size measurement based on optical vision as claimed in claim 4, is characterized in that: obtain the pixel coordinates of a plurality of light spots in the described S12, the concrete steps of calculating this light spot image perimeter L are as follows : S121: using the W spot images acquired in S11 as a training set; S122: Carry out geometric figure annotation on the W spot images respectively, and obtain W light spot images with graphic annotation labels, and each light spot image with graphic annotation labels is a training sample; S123: let t=1; S124: Select the tth training sample from W as the input of the Mask R-CNN neural network model, and output the Mask mask segmentation image t' corresponding to the tth training sample; S125: Perform contour fitting and edge detection processing on t' to obtain the predicted image t" of the tth training sample; S126: Set the loss threshold, and calculate the loss Loss between t" and t; When the Loss value is less than the loss threshold, the trained Mask R-CNN neural network model is obtained, and the next step is performed; otherwise, the parameters in the Mask R-CNN neural network model are updated by backpropagation, and t=t+1 is set, And return to S124; S127: Marking the image of the to-be-predicted light spot with a geometric figure, and using it as an input of the trained Mask R-CNN neural network model, and outputting the pixel coordinates of the light spot surrounding the image of the to-be-predicted light spot; S128: Calculate the perimeter L of the to-be-predicted spot image by using the spot pixel coordinates obtained in S127.

Images