Method and system for detecting persons and ships in lake

Abstract

The invention belongs to the technical field of target monitoring, and discloses a lake person and ship detection method and system. The method comprises the following steps: acquiring an image pair of a to-be-detected area of a lake, wherein the image pair comprises an infrared image and a visible light image of the to-be-detected area at the same time and under the same field of view; performing man-ship target detection on the two images in the same image pair through a neural network model to obtain a man-ship target preliminary detection result; based on the preliminary detection result of the man-ship target, calculating the intersection-to-union ratio of two image detection target position areas in the same image pair and the intersection-to-union ratio of detection target position areas between front and back frames of the image; and carrying out secondary judgment on the man-ship target based on the traffic and comparison. According to the method, false detection or false detection probability is reduced, and the robustness is extremely high; the influence of the environment on the detection result is reduced, and particularly, effective detection can be carried out at night or under severe weather conditions; the comprehensive monitoring of the lake area is realized; remote monitoring and early warning can be realized, and the real-time performance of detection is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of target monitoring, and in particular relates to a method and system for detecting people and boats on a lake. Background Art

[0002] As an important natural resource and tourist attraction, lakes not only have rich ecological value, but also high economic value. However, with the popularity of lake tourism and water sports, lake safety issues have become increasingly prominent.

[0003] Traditional lake safety management methods mainly rely on manual patrols and regular inspections. However, this method has the disadvantages of low efficiency, slow response, and potential safety hazards: manual patrols require a lot of manpower and material resources, especially when the lake area is large, it is difficult to achieve comprehensive monitoring coverage; manual patrols cannot achieve real-time monitoring, and once abnormal conditions are discovered, the best time to deal with them has often been missed; and in bad weather or at night, the safety and effectiveness of manual patrols are greatly reduced.

[0004] In order to solve the above problems, modern lake safety management gradually introduces cameras for remote monitoring. However, the existing lake monitoring technology still has the following shortcomings: (1) Affected by the environment: Existing lake monitoring cameras usually use only visible light or infrared light. Visible light cameras do not work well at night or in low light conditions, and although infrared cameras can work at night, they are easily disturbed by strong light during the day. Due to the complex lake environment, such as water reflections, weather changes and other factors, the image quality captured by a single camera is often not high, making it difficult to meet the requirements of accurate detection.

[0005] (2) Limited detection capability: Existing lake monitoring systems can only detect boats, and have a weaker ability to detect people, especially when the water surface is reflective or obstructed. For example, when the temperature is high at noon in summer, the temperature difference between people and trees on the ground is small, and infrared imaging has a weak ability to detect people on the lakeside.

[0006] (3) Insufficient real-time and accuracy: Due to the limitations of image processing and computer vision algorithms, the existing lake monitoring system still needs to be improved in terms of real-time and accuracy. Summary of the invention

[0007] In order to solve the above problems, one object of the present invention is to provide a method for detecting people and boats in a lake, which can reduce the interference of environmental factors on the detection results and improve the accuracy of detection.

[0008] The specific technical solutions are as follows: S1. Acquire an image pair of the lake area to be measured, wherein the image pair includes an infrared image and a visible light image of the area to be measured at the same time and in the same field of view; S2. Perform human-ship target detection on two images in the same image pair through a neural network model, output target detection sets of infrared images and visible light images respectively, and obtain preliminary detection results of human-ship targets; S3. Based on the preliminary detection results of the human-ship targets, calculate the intersection-and-union ratio of the detection target position areas of the two images in the same image pair and the intersection-and-union ratio of the detection target position areas between the previous and next frames of the image; based on the intersection-and-union ratio, output the detection set of the human-ship targets and perform a secondary judgment on the human-ship targets.

[0009] Furthermore, the step S3 comprises: S31. Based on the preliminary detection results of the human-ship target, determine the existence of the detection target in the same image pair of the current frame; if the detection target exists in both images of the same image pair, execute step S32; if the detection target exists in only one image of the same image pair, execute step S33; if the detection target does not exist in both images of the same image pair, execute step S1; S32. Map the detection target position area of ​​one image in the same image pair to the detection set obtained in the other image, recorded as the mapping detection set; based on the mapping detection set and the target detection set of the other image, determine whether the intersection and union ratio of the position areas of the same type of targets in the two images of the current frame is greater than or equal to a first threshold; if so, obtain the target detection set of the current frame image pair based on the target detection set of one image of the current frame and the corresponding mapping detection set, and proceed to step S34; if not, proceed to step S33; S33. One image in the same image pair in which the detection target is determined to exist is recorded as the first image, and the other image is recorded as the second image; calculate whether the intersection and union ratio of the detection target position area of ​​the first image of the current frame and the detection target position area in the detection set of the human-ship target of the previous frame is greater than or equal to the second threshold; if so, based on the mapping detection set obtained by mapping the target detection position area in the first image of the current frame to the second image and the detection set of the human-ship target of the previous frame, obtain the target detection set of the current frame image pair; and proceed to step S34; S34. Based on the motion changes of the target position area detected in the previous and next frames of the image, the detection set of the human-ship target in the current frame is output, and a secondary judgment is made on the current human-ship target.

[0010] In step S32, the image is used for the current frame target detection set U The calculation method is: ; in, The target detection location area of ​​one image in the current frame is mapped to the mapping detection set in another image in the same image pair, is the target detection set in another image of the same image pair in the current frame, , is the weight coefficient, and .

[0011] Furthermore, in step S33, the image is used for the current frame target detection set U The calculation method is: ; in, The target detection location area in the first image of the current frame is mapped to the corresponding mapping detection set in the second image, is the detection set of human-ship targets in the previous frame, Initialized to an empty set, , is the weight coefficient, and .

[0012] Furthermore, the method further includes setting a target counter, and performing a human-ship target detection alarm based on the target counter: In the step S31, when there is no detection target in both images, the target counter is reduced by one; In the step S32, when the intersection-and-union ratio of the position areas of the same type of objects in the two images is greater than or equal to the first threshold, the object counter is incremented by one; In the step S33, when the intersection-and-union ratio of the detection target position area of ​​the first image of the current frame and the detection target position area of ​​the detection set of the human-ship target of the previous frame is greater than or equal to the second threshold, the target counter is increased by one; otherwise, the target counter is decreased by one; In step S34, when the target counter reaches the target counter threshold, the target detection set of the current frame image pair is output as the detection set of the human-ship target.

[0013] Furthermore, in step S34, When the target counter does not reach the target counter threshold, the corresponding mapping detection set is assigned to the next frame , go to step S1 to make the next frame judgment; wherein, is the detection set of human-ship targets in the previous frame, Initialize to an empty set; When the target counter reaches the target counter threshold, the target detection set of the current frame image pair is output as the detection set of the human-ship target, and the target detection set of the current frame image pair is assigned to the next frame At the same time, the target counter is reduced by one, and the process goes to step S1 to make the next frame judgment.

[0014] Furthermore, a dual-light gimbal integrating dual infrared cameras and visible light cameras is used to collect infrared images and visible light images.

[0015] Furthermore, in step S1, infrared images and visible light images are collected by an infrared camera and a visible light camera; before step S1, the infrared camera and the visible light camera are calibrated, and the homography matrix from the infrared camera to the visible light camera and the homography matrix from the visible light camera to the infrared camera are calculated; in step S3, according to the homography matrix, the target detection position area of ​​one image in the image pair is mapped to the other image.

[0016] Furthermore, the calculation method of the homography matrix specifically includes: For a set of target feature points in the same field of view and world coordinate system , in the infrared and visible light target images, find two sets of feature points that correspond to one and ; According to the camera imaging model, the infrared camera has ,in , , They are the intrinsic matrix, rotation matrix, and translation matrix of the infrared camera respectively; for the visible light camera, ,in , , They are the internal parameter matrix, rotation matrix, and translation matrix of the visible light camera respectively; for the common target feature points Doing matrix transformation, we get: ;set up , we can get a one-to-one mapping from infrared to visible light images: ;in is the homography matrix between images, which can be solved by RANSCA or LMEDS algorithm when the number of feature points is greater than 3.

[0017] Another object of the present invention is to provide a lake person-boat detection system, which applies the above-mentioned lake person-boat detection method, and the system comprises: Visible light module, used to obtain visible light images of the lake area to be tested; Infrared core, used to obtain infrared images of the lake area to be measured; A network module, used for transmitting the visible light image and the infrared image to a processing module; The processing module is used to output the target position of people and ships and issue warnings based on visible light images and infrared images.

[0018] Compared with the prior art, one or more of the above technical solutions can achieve at least one of the following beneficial effects: This method is based on infrared light and visible light, combined with image fusion technology and optimized computer vision algorithm. By effectively making a secondary judgment on the detection target, it significantly reduces the probability of false detection or misdetection and has strong robustness. This method can avoid inaccurate detection results caused by missed detection or false detection in alternating visible light or infrared light images by optimizing program design; This method reduces the impact of the environment on the test results, and can be used to perform effective tests at night or under adverse weather conditions; This method realizes comprehensive monitoring of the lake area and improves safety; it can realize remote monitoring and early warning, and reduces the need for manual inspections; This method improves the real-time nature of detection and helps to quickly respond to emergency situations; This method realizes comprehensive, efficient and accurate monitoring of the lake area and effectively improves the level of lake safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0020] Figure 1 This is a flowchart of the lake person and boat detection method of Example 1.

[0021] Figure 2 This is a specific flow chart of the lake person and boat detection method of Example 1.

[0022] Figure 3 This is a schematic diagram of the composition of the lake person and boat detection system of Example 2.

[0023] Figure 4 This is Example 1 of the lake person and boat detection results of Example 2.

[0024] Figure 5 This is Example 2 of the lake person and boat detection results of Example 2. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] Example 1 like Figure 1 As shown, the present invention provides a method for detecting people and ships in a lake area, and the method includes the following steps: obtaining an image pair of the lake area to be tested, the image pair including an infrared image and a visible light image of the area to be tested at the same time and in the same field of view; performing human and ship target detection on the two images in the same image pair through a neural network model, and outputting target detection sets of infrared images and visible light images respectively, to obtain preliminary detection results of human and ship targets; based on the preliminary detection results of human and ship targets, calculating the intersection and union ratio of the detection target position areas of the two images in the same image pair, and the intersection and union ratio of the detection target position areas between the previous and next frames of the image; based on the intersection and union ratio, outputting the detection set of human and ship targets, and performing secondary judgment on human and ship targets. The method of this embodiment mainly captures infrared images and visible light images of the area to be tested at the same time, maps the target detection position area of ​​one of the images to the other image, and judges the detection target position area by the intersection and union ratio of the target detection position areas of the infrared and visible light images; and in the case where the target is not detected in a single frame, the relationship between the previous and next frames can be used for judgment, so as to achieve the purpose of regional target detection.

[0027] like Figure 2 As shown, as a specific implementation, the method for detecting people and boats in lake areas specifically includes the following steps: S0. Calibrate the visible light camera and the infrared camera, and calculate the homography matrix from the visible light camera to the infrared camera and the homography matrix from the infrared camera to the visible light camera.

[0028] The following takes the calculation of the homography matrix from an infrared camera to a visible light camera as an example.

[0029] For a set of target feature points in the same field of view and world coordinate system , in the infrared and visible light target images, find two sets of feature points that correspond to one and ; According to the camera imaging model, for infrared cameras: ,in , , They are the intrinsic matrix, rotation matrix, and translation matrix of the infrared camera respectively; similarly, the visible light camera has: ,in , , They are the internal parameter matrix, rotation matrix and translation matrix of the visible light camera respectively; Doing matrix transformation, we get: ;set up , we can get a one-to-one mapping from infrared to visible light images: ,in is the homography matrix between images, which can be solved by RANSCA or LMEDS algorithm when the number of feature points is greater than 3 (non-collinear).

[0030] After calibrating the infrared and visible light cameras, two neural network models are set before detecting people and ships. They are called neural network Yolo v5A and Yolo v5B, which are used to detect people, ships, cars and other targets based on visible light images and infrared images respectively. The neural network is initialized with infrared and visible light network parameters respectively. Before detecting people and ships, the neural network model is trained first; the trained neural networks Yolo v5A and Yolo v5B are loaded with their own trained network weight coefficients respectively. The method also sets a target counter, which is used for human-ship target detection and alarm based on the target counter in the subsequent steps. Before human-ship target detection, initialize the target counter and target counter threshold ,set up .

[0031] S1. Obtain an image pair of the lake area to be measured, wherein the image pair includes an infrared image of the area to be measured and a corresponding visible light image at the same time and in the same field of view.

[0032] In the specific implementation, an infrared camera is used to obtain infrared images, and a visible light camera is used to obtain visible light images. As a preferred implementation, a dual-light gimbal that integrates an infrared camera and a visible light camera can be used to collect images.

[0033] S2. Use the neural network model to perform human-ship target detection on the two images in the same image pair respectively to obtain preliminary detection results of human-ship targets.

[0034] As a specific implementation, the type of neural network model in this embodiment adopts the Yolo v5s neural network model. It is understandable that other neural network models of the Yolo series can also be used. The neural network model can adopt a conventional structure, which is not repeated here.

[0035] The visible light image of the current frame is input into the neural network Yolo v5A, and the corresponding infrared light image of the current frame is input into the neural network Yolo v5B.

[0036] Suppose the output target detection set in the neural network Yolo v5A is , as the result of human-vessel detection based on visible light images; let the output target detection set in Yolo v5B be , as the human-ship detection result based on infrared light image. The human-ship detection result based on infrared light image and the human-ship detection result based on visible light image are taken as the preliminary results of human-ship target detection.

[0037] S3. Based on the preliminary detection results of the human-ship target, calculate the intersection-and-union ratio of the detection target position area of ​​the two images in the same image pair and the intersection-and-union ratio of the detection target position area between the previous and next frames of the image, output the detection set of the human-ship target based on the intersection-and-union ratio, and perform a secondary judgment on the human-ship target. Specifically, step S3 includes steps S31 to S34.

[0038] S31. Mapping judgment: Based on the preliminary detection results of the human-ship target, judge the existence of the detection target in the same image pair of the current frame; if the detection target exists in both images of the same image pair, that is , then execute step S32; if only one of the images in the same image pair has a detection target, , then execute step S33; if there is no detection target in both images, then When , the target counter is reduced by one and the process returns to step S1.

[0039] S32. Bi-optical intersection-and-union ratio: map the detection target position areas of one of the images in the same image pair to the other image one by one, and record the obtained detection set as the mapping detection set; based on the mapping detection set and the target detection set of the other image, determine whether the intersection-and-union ratio of the position areas of the same type of targets in the two images is greater than or equal to a first threshold; if so, calculate the target detection set of the current frame image pair based on the target detection set of one image in the current frame and the corresponding mapping detection set, and accumulate the target counter, and proceed to step S34; if not, proceed to step S33.

[0040] Among them, the calculation method of the target detection set U for the current frame image is: ; in, The target detection location area of ​​one image in the current frame is mapped to the mapping detection set in another image in the same image pair, is the target detection set in another image of the same image pair in the current frame, , is the weight coefficient, and .

[0041] In specific implementation, the target detection position area of ​​one image in the same image pair can be mapped to the other image according to the situation: for example, the target detection position area of ​​the infrared image can be mapped to the visible light image, and the target detection position area of ​​the visible light image can also be mapped to the corresponding infrared image.

[0042] In this embodiment, the homography matrix is ​​used The target detection set The detection target location area in Mapped one by one to the visible light image, the target detection set in the infrared image The set of mapping detections in the visible light image is set to ,in .

[0043] calculate and In the The intersection of the detected target location area : ,in for In the Detect target location area. Set The threshold is the first threshold ,when When the target detection set of the current frame image pair is ,in , and the current target counter , and proceed to step S34; otherwise, it is determined that the detection targets of the two images are not the same target, and the process proceeds to step S33 for inter-frame intersection and comparison judgment.

[0044] Similarly, if the location area of ​​the target detection in the visible light image is mapped to the infrared image, the above method can be referred to to obtain the target detection set of the current frame image pair based on the mapping detection set of the current frame visible light image mapped to the infrared image and the target detection set of the current frame infrared image.

[0045] S33. Inter-frame intersection and union ratio: record one image in the same image pair where the detection target is judged to exist as the first image, and record the other image as the second image; calculate whether the intersection and union ratio of the detection target position area of ​​the current frame of the first image and the detection target position area in the detection set of the human and ship targets in the previous frame is greater than or equal to the second threshold; if so, obtain the target detection set of the current frame image pair based on the mapping detection set obtained by mapping the target detection position area in the first image of the current frame to the second image and the detection set of the human and ship targets in the previous frame; proceed to step S34. Among them, when there are detection targets in both images of the image pair, one of them can be selected as the first image and the other image as the second image; when there is a detection target in only one of the two images of the image pair, the image with the detection target can be used as the first image and the other image without the detection target can be used as the second image. The calculation method of the current frame target detection set U of the image pair is: ; in, The target detection location area in the first image of the current frame is mapped to the corresponding mapping detection set in the second image, is the detection set of human-ship targets in the previous frame, Initialized to an empty set, , is the weight coefficient, and .

[0046] In the following, an example is taken in which the infrared image in the image pair has a detection target, but the visible light image does not have a detection target, that is, the first image is an infrared image, and the second image is a visible light image.

[0047] Specifically, when there is only one undetected target in infrared or visible light (with As an example), use the homography matrix The target detection set The detection target location area in Mapped one by one to the visible light image, the target detection set in the infrared image The set of mapping detections in the visible light image is set to . Assume that the detection set of the human-ship target in the previous frame is , refer to the same calculation method as above, calculate and In the The intersection of the detected target location area .set up The threshold is the second threshold ,when When the target detection set of the current frame image pair is ,in , and the current target counter ,otherwise ; Execute step S34.

[0048] S34. Area judgment: According to the movement changes of the target position area detected in the previous and next frames of the image, the detection set of the human and ship targets in the current frame is output; the position of the current target point is calculated, and the current human and ship targets are secondarily judged.

[0049] Determine the current target counter Whether the target counter threshold is reached , if the target counter threshold is reached , then the target detection set of the current frame image pair U As the detection set of human and ship targets, the current target point position output U out ， This is the secondary judgment result.

[0050] And, if the current target counter Target counter threshold reached , output the target detection set of the current frame image pair At the same time, the target detection set of the current frame image pair Assign to the next frame , that is, ,at the same time ;like If it is not satisfied, then assign the corresponding mapping detection set to the next frame ,Right now ; Return to step S1 to make the next frame judgment.

[0051] This embodiment is based on the dual-light gimbal to detect and determine people and ships in the lake area. According to the detection results of the current infrared and visible light images, the visible light image and the infrared image can be mapped by homography matrix to combine and jointly determine the detection results of specific targets. It can avoid missed detection and false detection caused by changes in external conditions (such as illumination changes, target occlusion) in the infrared or visible light single image, and especially avoid the problem of false detection and missed detection in the infrared image and visible light image alternately. In addition, when there is a situation where the other camera cannot detect normally, such as at night (the visible light camera cannot obtain effective images) or outside the long-distance field of view (the visible light field of view is large and the distance is far), the regional human and ship detection method of the dual-light gimbal can be degenerated into a single target detection of the infrared or visible light camera; therefore, this method also has strong environmental adaptability. This method sets a count to realize the filtering function, and filters out most of the occasional false detections by counting continuous frames.

[0052] Example 2 like Figure 3 As shown, this embodiment provides a lake person-and-vessel detection system, which applies the lake person-and-vessel detection method of Embodiment 1. The system includes: a visible light module, which is used to obtain a visible light image of the lake area to be tested; an infrared core, which is used to obtain an infrared image of the lake area to be tested; a network module, which is used to transmit the visible light image and the infrared image to the processing module; and the processing module, which is used to output the target position of the person-and-vessel according to the visible light image and the infrared image and issue a warning.

[0053] As a specific implementation, the parameters of the visible light module, infrared movement, and network module are shown in Table 1, Table 2, and Table 3.

[0054] Table 1 Visible light module parameters Table 2 Infrared movement parameters Table 3 Network module parameters In the specific implementation process, visible light modules and infrared cores are deployed around the lake and connected to the monitoring center through a wireless network. The processing module is the server of the monitoring center, on which image processing and computer vision algorithms are run to analyze the real-time transmitted images. The processing module can be logged in remotely to play the role of real-time monitoring and processing warning information.

[0055] When an abnormal situation is detected, the system automatically sends an early warning message to the monitoring personnel and records relevant images and data. Figure 4 The following is an example of using the system to detect lake boats: Figure 5 The following is an example of using this system to detect people in a lake. The monitoring personnel make an assessment based on the early warning information and take appropriate emergency measures.

[0056] The detection system of this embodiment realizes remote monitoring and early warning, reduces the need for manual inspections, improves the real-time performance of detection, and helps to quickly respond to emergencies.

[0057] Obviously, the above embodiments are merely examples for clearly explaining the technical solutions of the present invention, and are not intended to limit the implementation methods of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the protection of the claims of the present invention.

Claims

1. A method for detecting people and boats on a lake, characterized in that: The following steps are involved: S1. Acquire an image pair of the lake area to be measured, wherein the image pair includes an infrared image and a visible light image of the area to be measured at the same time and in the same field of view; S2. Perform human-ship target detection on two images in the same image pair through a neural network model, output target detection sets of infrared images and visible light images respectively, and obtain preliminary detection results of human-ship targets; S3. Based on the preliminary detection results of the human-ship targets, calculate the intersection-and-union ratio of the detection target position areas of the two images in the same image pair and the intersection-and-union ratio of the detection target position areas between the previous and next frames of the image; based on the intersection-and-union ratio, output the detection set of the human-ship targets and perform a secondary judgment on the human-ship targets.

2. The lake person-boat detection method according to claim 1, characterized in that: The step S3 comprises: S31. Based on the preliminary detection results of the human-ship target, determine the existence of the detection target in the same image pair of the current frame; if the detection target exists in both images of the same image pair, execute step S32; if the detection target exists in only one image of the same image pair, execute step S33; if the detection target does not exist in both images of the same image pair, execute step S1; S32. Map the detection target position area of ​​one image in the same image pair to the detection set obtained in the other image, recorded as the mapping detection set; based on the mapping detection set and the target detection set of the other image, determine whether the intersection and union ratio of the position areas of the same type of targets in the two images of the current frame is greater than or equal to a first threshold; if so, obtain the target detection set of the current frame image pair based on the target detection set of one image of the current frame and the corresponding mapping detection set, and proceed to step S34; if not, proceed to step S33; S33. One image in the same image pair in which the detection target is determined to exist is recorded as the first image, and the other image is recorded as the second image; calculate whether the intersection and union ratio of the detection target position area of ​​the first image of the current frame and the detection target position area in the detection set of the human-ship target of the previous frame is greater than or equal to the second threshold; if so, based on the mapping detection set obtained by mapping the target detection position area in the first image of the current frame to the second image and the detection set of the human-ship target of the previous frame, obtain the target detection set of the current frame image pair; and proceed to step S34; S34. Based on the motion changes of the target position area detected in the previous and next frames of the image, the detection set of the human-ship target in the current frame is output, and a secondary judgment is made on the current human-ship target.

3. The lake person-boat detection method according to claim 2, characterized in that: In step S32, the image is used for the current frame target detection set U The calculation method is: ； in, The target detection location area of ​​one image in the current frame is mapped to the mapping detection set in another image in the same image pair, is the target detection set in another image of the same image pair in the current frame, , is the weight coefficient, and .

4. The lake person-boat detection method according to claim 2, characterized in that: In step S33, the image is used for the current frame target detection set U The calculation method is: ； in, The target detection location area in the first image of the current frame is mapped to the corresponding mapping detection set in the second image, is the detection set of human-ship targets in the previous frame, Initialized to an empty set, , is the weight coefficient, and .

5. The lake person-vessel detection method according to any one of claims 2 to 4, characterized in that: The method further includes setting a target counter, and performing a human-ship target detection alarm based on the target counter: In the step S31, when there is no detection target in both images, the target counter is reduced by one; In the step S32, when the intersection-and-union ratio of the position areas of the same type of objects in the two images is greater than or equal to the first threshold, the object counter is incremented by one; In the step S33, when the intersection-and-union ratio of the detection target position area of ​​the first image of the current frame and the detection target position area of ​​the detection set of the human-ship target of the previous frame is greater than or equal to the second threshold, the target counter is increased by one; otherwise, the target counter is decreased by one; In step S34, when the target counter reaches the target counter threshold, the target detection set of the current frame image pair is output as the detection set of the human-ship target.

6. The lake person-boat detection method according to claim 5, characterized in that: In the step S34, When the target counter does not reach the target counter threshold, the corresponding mapping detection set is assigned to the next frame , go to step S1 to make the next frame judgment; wherein, is the detection set of human-ship targets in the previous frame, Initialize to an empty set; When the target counter reaches the target counter threshold, the target detection set of the current frame image pair is output as the detection set of the human-ship target, and the target detection set of the current frame image pair is assigned to the next frame At the same time, the target counter is reduced by one, and the process goes to step S1 to make the next frame judgment.

7. The lake person-boat detection method according to claim 1, characterized in that: A dual-light gimbal that integrates dual infrared cameras and visible light cameras is used to collect infrared images and visible light images.

8. The lake person-boat detection method according to claim 2, characterized in that: In step S1, infrared images and visible light images are collected by an infrared camera and a visible light camera; before step S1, the infrared camera and the visible light camera are calibrated, and the homography matrix from the infrared camera to the visible light camera and the homography matrix from the visible light camera to the infrared camera are calculated; in step S3, according to the homography matrix, the target detection position area of ​​one image in the image pair is mapped to the other image.

9. The lake person-boat detection method according to claim 8, characterized in that: The calculation method of the homography matrix specifically includes: For a set of target feature points in the same field of view and world coordinate system , in the infrared and visible light target images, find two sets of feature points that correspond to one and ; According to the camera imaging model, the infrared camera has ,in , , They are the intrinsic matrix, rotation matrix, and translation matrix of the infrared camera respectively; for the visible light camera, ,in , , They are the internal parameter matrix, rotation matrix, and translation matrix of the visible light camera respectively; for the common target feature points Doing matrix transformation, we get: ;set up , we can get a one-to-one mapping from infrared to visible light images: ;in is the homography matrix between images, which can be solved by RANSCA or LMEDS algorithm when the number of feature points is greater than 3.

10. A lake person and boat detection system, characterized in that: The lake person-boat detection method according to any one of claims 1 to 9 is applied, and the system comprises: Visible light module, used to obtain visible light images of the lake area to be tested; Infrared core, used to obtain infrared images of the lake area to be measured; A network module, used for transmitting the visible light image and the infrared image to a processing module; The processing module is used to output the target position of people and ships and issue warnings based on visible light images and infrared images.

Images