Non-motor vehicle manned and overload detection system and method based on road monitoring

Abstract

The invention discloses a non-motor vehicle manned and overload detection system and method based on road monitoring. The non-motor vehicle manned and overload detection system based on road monitoring comprises a data acquisition module and an image preprocessing module, the data acquisition module extracts pictures from monitoring equipment or imports the pictures into the data acquisition module from the monitoring equipment so as to output a first intermediate data set; and the image preprocessing module is used for acquiring the first intermediate data set and carrying out standardized image processing and contrast enhancement processing on the first intermediate data set so as to output a second intermediate data set. The non-motor vehicle manned and overload detection system and method based on road monitoring have the advantages that road monitoring processing scenes of pictures and videos are adapted, and meanwhile real-time monitoring video analysis and overload detection of off-line pictures are supported.

Description

Technical Field

[0001] The present invention belongs to the field of non-motor vehicle overload detection, and in particular relates to a non-motor vehicle passenger and overload detection system and method based on road monitoring. Background Art

[0002] The publication number is CN114394179A, the subject name is invention patent application for non-motor vehicle illegal passenger carrying detection method and electronic equipment, and its IPC classification number is B62J45 / 41. Its technical solution discloses "obtaining multiple pressure trigger signals, the multiple pressure trigger signals are respectively detected by multiple pressure sensors arranged on the seating positions of the vehicle; judging whether it is non-compliant riding or compliant riding based on the pressure trigger signals detected by the multiple pressure sensors, and obtaining a detection result; the multiple pressure sensors are arranged front and back in sequence on the seating positions and numbered sequentially, each of the pressure sensors having a serial number; judging whether it is non-compliant riding or compliant riding based on the pressure trigger signals detected by the multiple pressure sensors".

[0003] It can be seen that the above invention patent application has disclosed one of the technical solutions for the non-motor vehicle passenger and overload detection method. However, the technical solution disclosed in the above invention patent application has certain limitations in its scope of application. First, it is limited to judging violations through pressure signals and does not involve a wider range of road monitoring scenarios. Secondly, its support capabilities for real-time monitoring video analysis and offline image processing in complex traffic environments are relatively weak. In addition, the existing technology still needs to be further improved for the accurate identification of non-motor vehicles and the number of passengers. Summary of the invention

[0004] In view of the status of the prior art, the present invention overcomes the above-mentioned defects and provides a non-motor vehicle passenger and overload detection system and method based on road monitoring, improves the existing non-motor vehicle passenger and overload detection technology, especially focusing on the real-time performance of road monitoring, the compatibility of offline image processing and the adaptability to complex environments.

[0005] The present invention adopts the following technical scheme, a non-motor vehicle passenger and overload detection system based on road monitoring, including:

[0006] A data acquisition module, which extracts pictures from the monitoring device or imports pictures from the monitoring device to the data acquisition module to output a first intermediate data set;

[0007] An image preprocessing module, wherein the image preprocessing module acquires a first intermediate data set, and performs standardized image processing and contrast enhancement processing on the first intermediate data set to output a second intermediate data set;

[0008] A target detection module, wherein the target detection module obtains a second intermediate data set, and the target detection module imports the second intermediate data set into a preset deep learning module to output non-motor vehicle and personnel targets;

[0009] A people counting module, which obtains non-motor vehicles and human targets, and combines the non-motor vehicles and human targets with spatial analysis to output the number of people;

[0010] The overload determination module obtains the number of personnel, determines whether the number of personnel is overloaded according to a preset threshold, and performs corresponding actions.

[0011] As a preferred technical solution of the above technical solution, the data acquisition module includes a video processing unit and a picture processing unit, wherein:

[0012] The video processing unit operates at a fixed frame rate F rate The proposed frame sequence {f i};

[0013] The image processing unit directly imports the image collection {I k}, the single picture number is k.

[0014] As a preferred technical solution of the above technical solution, the image preprocessing module includes a size standardization unit and an enhancement operation unit, wherein:

[0015] Size normalization unit will be picture I k The size is adjusted to S in , S in Characterized by formula (1):

[0016] S in =640×640 (1);

[0017] The image after size standardization is denoted as I′ k , I′ k Characterized by formula (2):

[0018] I′ k =Resize(I k , S in ) (2);

[0019] The enhancement operation unit optimizes the image quality by using a contrast enhancement algorithm (preferably using a CLAHE algorithm);

[0020] The image after image quality optimization is denoted as I″ k , I″ k Characterized by formula (3):

[0021] I″ k =CLAHE(I′k ) (3).

[0022] As a preferred technical solution of the above technical solution, the target detection module includes a target bounding box detection unit and a category label unit, wherein:

[0023] The target bounding box detection unit processes the image I″ after image quality optimization k , to output the target bounding box b ij , target bounding box b ij It is represented by formula (4):

[0024] b ij ={(x 1 ,y 1 , x 2 ,y 2 ), c, s} (4);

[0025] Among them, x 1 ,y 1 is the target bounding box b ij The upper left corner coordinate of 2 ,y 2 is the target bounding box b ij The lower right corner coordinate of the target; c is the classification confidence of the target; s is the target bounding box b ij area;

[0026] The category label unit processes the target bounding box b ij , to output non-motor vehicle category labels and personnel category labels;

[0027] Single image target detection, use the YOLOv5 model or the YOLOv7 model to perform target detection to generate a set of bounding boxes B for each target img , the bounding box set B img It is represented by formula (5):

[0028] B img = {b ij |j∈[1,N]} (5);

[0029] Among them, B img is the target detection result set of the image, and N is the total number of detected targets.

[0030] As a preferred technical solution of the above technical solution, the personnel counting module is specifically implemented as the following technical solution:

[0031] Counting people in the image, by using the spatial relationship of the bounding box, counting the number of people N in each non-motor vehicle person (k), N person (k) is represented by formula (6):

[0032]

[0033] Among them, IoU(b ij , B motor ) is the overlap rate, and the value of δ(x) is as follows: when x>T iou When δ(x) is 1, otherwise δ(x) is 0;

[0034] The overload determination module is specifically implemented as the following technical solutions:

[0035] The overload determination method for images and videos combines the non-motor vehicle information detected in each image or video frame with the number of people. The statistically combined overload sign is recorded as Overload(k). Overload(k) is represented by formula (7):

[0036]

[0037] Where Overload(k) is the overload flag of picture k or video frame k, T max is the passenger threshold for non-motor vehicles;

[0038] Directly process the partition analysis method of the image. The distribution of non-motor vehicles and people in the image is analyzed by partition, and the partition number is recorded as D zone (x,y),D zone (x, y) is represented by formula (8):

[0039] D zone (x, y) = ZoneID((x 1 +x 2 ) / 2,(y 1 +y 2 ) / 2) (8);

[0040] Among them, D zone (x, y) is the partition number where the target (x, y) is located, and the partition matrix is ​​dynamically divided according to the image size.

[0041] The present invention adopts the following technical scheme, and the non-motor vehicle carrying and overloading detection method based on road monitoring includes the following steps:

[0042] Step S1: the data acquisition module extracts pictures from the monitoring device or imports pictures from the monitoring device to the data acquisition module to output a first intermediate data set;

[0043] Step S2: the image preprocessing module acquires a first intermediate data set, and performs standardized image processing and contrast enhancement processing on the first intermediate data set to output a second intermediate data set;

[0044] Step S3: the target detection module acquires a second intermediate data set, and the target detection module imports the second intermediate data set into a preset deep learning module to output non-motor vehicle and human targets;

[0045] Step S4: the people counting module obtains non-motor vehicles and people targets, and the people counting module combines the non-motor vehicles and people targets with the spatial analysis to output the number of people;

[0046] Step S5: The overload determination module obtains the number of personnel, determines whether the number of personnel is overloaded according to a preset threshold, and performs corresponding actions.

[0047] As a preferred technical solution of the above technical solution, the data acquisition module includes a video processing unit and a picture processing unit, wherein:

[0048] The video processing unit operates at a fixed frame rate F rate The proposed frame sequence {f i};

[0049] The image processing unit directly imports the image collection {I k}, the single picture number is k.

[0050] As a preferred technical solution of the above technical solution, the image preprocessing module includes a size standardization unit and an enhancement operation unit, wherein:

[0051] Size normalization unit will be picture I k The size is adjusted to S in , S in Characterized by formula (1):

[0052] S in =640×640 (1);

[0053] The image after size standardization is denoted as I′ k , I′ k Characterized by formula (2):

[0054] I′ k =Resize(I k , S in ) (2);

[0055] The enhancement operation unit optimizes the image quality by using a contrast enhancement algorithm (preferably using a CLAHE algorithm);

[0056] The image after image quality optimization is denoted as I″ k , I″ k Characterized by formula (3):

[0057] I″ k =CLAHE(I′ k) (3).

[0058] As a preferred technical solution of the above technical solution, the target detection module includes a target bounding box detection unit and a category label unit, wherein:

[0059] The target bounding box detection unit processes the image I″ after image quality optimization k , to output the target bounding box b ij , target bounding box b ij Characterized by formula (4):

[0060] b ij ={(x 1 ,y 1 , x 2 ,y 2 ), c, s} (4);

[0061] Among them, x 1 ,y 1 is the target bounding box b ij The upper left corner coordinate of 2 ,y 2 is the target bounding box b ij The lower right corner coordinate of the target; c is the classification confidence of the target; s is the target bounding box b ij area;

[0062] The category label unit processes the target bounding box b ij , to output non-motor vehicle category labels and personnel category labels;

[0063] Single image target detection, use the YOLOv5 model or the YOLOv7 model to perform target detection to generate a set of bounding boxes B for each target img , the bounding box set B img It is represented by formula (5):

[0064] B img = {b ij |j∈[1,N]} (5);

[0065] Among them, B img is the target detection result set of the image, and N is the total number of detected targets.

[0066] As a preferred technical solution of the above technical solution, the personnel counting module is specifically implemented as the following technical solution:

[0067] Counting people in the image, by using the spatial relationship of the bounding box, counting the number of people N in each non-motor vehicle person (k), N person (k) is represented by formula (6):

[0068]

[0069] Among them, IoU(b ij , B motor ) is the overlap rate, and the value of δ(x) is as follows: when x>T iou When δ(x) is 1, otherwise δ(x) is 0;

[0070] The overload determination module is specifically implemented as the following technical solutions:

[0071] The overload determination method for images and videos combines the non-motor vehicle information detected in each image or video frame with the number of people. The statistically combined overload sign is recorded as Overload(k). Overload(k) is represented by formula (7):

[0072]

[0073] Where Overload(k) is the overload flag of picture k or video frame k, M max is the passenger threshold for non-motor vehicles;

[0074] Directly process the partition analysis method of the image. The distribution of non-motor vehicles and people in the image is analyzed by partition, and the partition number is recorded as D zone (x,y),D zone (x, y) is represented by formula (8):

[0075] D zone (x, y) = ZoneID((x 1 +x 2 ) / 2,(y 1 +y 2 ) / 2) (8);

[0076] Among them, D zone (x, y) is the partition number where the target (x, y) is located, and the partition matrix is ​​dynamically divided according to the image size.

[0077] The non-motor vehicle passenger and overload detection system and method based on road monitoring disclosed in the present invention has the following beneficial effects:

[0078] 1. Adapt to road monitoring processing scenarios of pictures and videos, and support real-time analysis of monitoring videos and overload detection of offline pictures.

[0079] 2. High efficiency and precision, combining the YOLO model and spatial analysis to achieve accurate detection in complex environments.

[0080] 3. High flexibility, suitable for various application scenarios such as traffic management, urban safety supervision, etc. DETAILED DESCRIPTION

[0081] The present invention discloses a non-motor vehicle passenger and overload detection system and method based on road monitoring. The specific implementation of the present invention is further described below in conjunction with a preferred embodiment (Example 1).

[0082] Those skilled in the art should note that the “picture”, “image”, “frame” and “video frame” that may be involved in various embodiments of the present invention are the same concept and are no longer distinguished.

[0083] Example 1.

[0084] Preferably, the non-motor vehicle passenger and overload detection system based on road monitoring includes:

[0085] A data acquisition module, which extracts pictures (for example, video frames of a video) from the monitoring device or imports pictures (video frames of a video) from the monitoring device to the data acquisition module to output a first intermediate data set that can be input into the image preprocessing module;

[0086] An image preprocessing module, wherein the image preprocessing module obtains a first intermediate data set, and performs standardized image processing and contrast enhancement processing on the first intermediate data set to output a second intermediate data set that can be input into the target detection module;

[0087] A target detection module, wherein the target detection module obtains a second intermediate data set, and the target detection module imports the second intermediate data set into a preset deep learning module to output non-motor vehicle and human targets that can be input into a human counting module;

[0088] A people counting module, which obtains non-motor vehicles and people targets, and combines the non-motor vehicles and people targets with spatial analysis to output the number of people that can be input into the overload determination module;

[0089] The overload determination module obtains the number of personnel, determines whether the number of personnel is overloaded according to a preset threshold, and performs corresponding actions.

[0090] The data acquisition module includes a video processing unit and an image processing unit, wherein:

[0091] The video processing unit operates at a fixed frame rate F rate The proposed frame sequence {f i};

[0092] The image processing unit directly imports the image collection {I k}, the single picture number is k.

[0093] The image preprocessing module includes a size standardization unit and an enhancement operation unit, wherein:

[0094] Size normalization unit will be picture I k The size is adjusted to S in , S in Characterized by formula (1):

[0095] S in =640×640 (1);

[0096] The image after size standardization is denoted as I′ k , I′ k Characterized by formula (2):

[0097] I′ k =Resize(I k , S in ) (2);

[0098] The enhancement operation unit optimizes the image quality by using a contrast enhancement algorithm (preferably using a CLAHE algorithm);

[0099] The image after image quality optimization is denoted as I″ k , I″ k Characterized by formula (3):

[0100] I″ k =CLAHE(I′ k ) (3).

[0101] The target detection module includes a target bounding box detection unit and a category label unit, where:

[0102] The target bounding box detection unit processes the image I″ after image quality optimization k , to output the target bounding box b ij , target bounding box b ij It is represented by formula (4):

[0103] b ij ={(x 1 ,y 1 , x 2 ,y 2 ), c, s} (4);

[0104] Among them, x 1 ,y 1 is the target bounding box b ij The upper left corner coordinate of 2 ,y 2 is the target bounding box b ij The lower right corner coordinate of the target; c is the classification confidence of the target; s is the target bounding box b ij The area (that is, size) of .

[0105] The category label unit processes the target bounding box b ij , to output non-motor vehicle category label (ID=3) and personnel category label (ID=0).

[0106] Among them, the target detection module can also be specifically implemented as the following technical solutions:

[0107] Single image target detection, use the YOLOv5 model or the YOLOv7 model to perform target detection to generate a set of bounding boxes B for each target img , the bounding box set B img It is represented by formula (5):

[0108] B img = {b ij |j∈[1,N]} (5);

[0109] Among them, B img is the target detection result set of the image, and N is the total number of detected targets.

[0110] Among them, the people counting module can also be specifically implemented as the following technical solutions:

[0111] Counting people in the image, by using the spatial relationship of the bounding box, counting the number of people N in each non-motor vehicle person (k), N person (k) is represented by formula (6):

[0112]

[0113] Among them, IoU(b ij , B motor ) is the overlap rate, and the value of δ(x) is as follows: when x>T iou When δ(x) is 1, otherwise δ(x) is 0.

[0114] Among them, the overload determination module can also be specifically implemented as the following technical solutions:

[0115] The overload determination method for images and videos combines the non-motor vehicle information detected in each image or video frame with the number of people. The statistically combined overload sign is recorded as Overload(k). Overload(k) is represented by formula (7):

[0116]

[0117] Where Overload(k) is the overload flag of picture k or video frame k, T max It is the passenger threshold of non-motor vehicles.

[0118] Among them, the overload determination module can also be specifically implemented as the following technical solutions:

[0119] Directly process the partition analysis method of the image. The distribution of non-motor vehicles and people in the image is analyzed by partition, and the partition number is recorded as D zone (x, y).D zone (x, y) is represented by formula (8):

[0120] D zone (x, y) = ZoneID((x 1 +x 2 ) / 2,(y 1 +y 2 ) / 2) (8);

[0121] Among them, D zone (x, y) is the partition number where the target (x, y) is located, and the partition matrix is ​​dynamically divided according to the image size.

[0122] Among them, the non-motor vehicle passenger and overload detection system based on road monitoring also includes a data storage and visualization module. In order to improve the comprehensive performance of detection, the data storage and visualization module outputs the image and video processing results in a unified JSON format. The image and video processing results include the image ID k, the number of non-motor vehicles M img (k) Average number of passengers And overload flag Overload(k).

[0123] Among them, the overload determination module can also be specifically implemented as the following technical solutions:

[0124] For the overload detection results of a single image, the cumulative overload probability P is predicted by the weighted probability model overload (k), P overload (k) is represented by formula (9):

[0125]

[0126] Among them, w ij is the weight of each target in the image, and δ(Overload(j)) is the overload status flag function.

[0127] Among them, the overload determination module can also be specifically implemented as the following technical solutions:

[0128] Combined with the characteristics of the partition, the statistical model of non-motor vehicles in the whole area is used to calculate the average passenger rate R of the partition. region , R region It is represented by formula (10):

[0129]

[0130] Example 2.

[0131] Preferably, the non-motor vehicle passenger and overload detection method based on road monitoring comprises the following steps:

[0132] Step S1: The data acquisition module extracts pictures (for example, video frames of a video) from the monitoring device or imports pictures (video frames of a video) from the monitoring device to the data acquisition module, so as to output a first intermediate data set that can be input into the image preprocessing module;

[0133] Step S2: the image preprocessing module acquires a first intermediate data set, and performs standardized image processing and contrast enhancement processing on the first intermediate data set to output a second intermediate data set that can be input into the target detection module;

[0134] Step S3: the target detection module obtains the second intermediate data set, and the target detection module imports the second intermediate data set into a preset deep learning module to output non-motor vehicle and human targets that can be input into the human counting module;

[0135] Step S4: The personnel counting module obtains non-motor vehicles and personnel targets, and the personnel counting module combines the non-motor vehicles and personnel targets with the spatial analysis to output the number of personnel that can be input into the overload determination module;

[0136] Step S5: The overload determination module obtains the number of personnel, determines whether the number of personnel is overloaded according to a preset threshold, and performs corresponding actions.

[0137] Wherein, in step S1, the data acquisition module includes a video processing unit and an image processing unit, wherein:

[0138] The video processing unit operates at a fixed frame rate F rate The proposed frame sequence {f i};

[0139] The image processing unit directly imports the image collection {I k}, the single picture number is k.

[0140] In step S2, the image preprocessing module includes a size standardization unit and an enhancement operation unit, wherein:

[0141] Size normalization unit will be picture I k The size is adjusted to S in , S in Characterized by formula (1):

[0142] S in =640×640 (1);

[0143] The image after size standardization is denoted as I′ k , I′ kCharacterized by formula (2):

[0144] I′ k =Resize(I k , S in ) (2);

[0145] The enhancement operation unit optimizes the image quality by using a contrast enhancement algorithm (preferably using a CLAHE algorithm);

[0146] The image after image quality optimization is denoted as I″ k , I″ k Characterized by formula (3):

[0147] I″ k =CLAHE(I′ k ) (3).

[0148] In step S3, the target detection module includes a target bounding box detection unit and a category label unit, wherein:

[0149] The target bounding box detection unit processes the image I″ after image quality optimization k , to output the target bounding box b ij , target bounding box b ij Characterized by formula (4):

[0150] b ij ={(x 1 ,y 1 , x 2 ,y 2 ), c, s} (4);

[0151] Among them, x 1 ,y 1 is the target bounding box b ij The upper left corner coordinate of 2 ,y 2 is the target bounding box b ij The lower right corner coordinate of the target; c is the classification confidence of the target; s is the target bounding box b ij The area (i.e. size) of .

[0152] The category label unit processes the target bounding box b ij , to output non-motor vehicle category label (ID=3) and personnel category label (ID=0).

[0153] Among them, in step S3, the target detection module can also be specifically implemented as the following technical solutions:

[0154] Single image target detection, use the YOLOv5 model or the YOLOv7 model to perform target detection to generate a set of bounding boxes B for each target img, the bounding box set B img It is represented by formula (5);

[0155] B img = {b ij |j∈[1,N]} (5);

[0156] Among them, B img is the target detection result set of the image, and N is the total number of detected targets.

[0157] Among them, in step S4, the people counting module can also be specifically implemented as the following technical solutions:

[0158] Counting people in the image, by using the spatial relationship of the bounding box, counting the number of people N in each non-motor vehicle person (k), N person (k) is represented by formula (6):

[0159]

[0160] Among them, IoU(b ij , B moror ) is the overlap rate, and the value of δ(x) is as follows: when x>T iou When δ(x) is 1, otherwise δ(x) is 0.

[0161] Among them, in step S5, the overload determination module can also be specifically implemented as the following technical solutions:

[0162] The overload determination method for images and videos combines the non-motor vehicle information detected in each image or video frame with the number of people. The statistically combined overload sign is recorded as Overload(k). Overload(k) is represented by formula (7):

[0163]

[0164] Where Overload(k) is the overload flag of picture k or video frame k, T max It is the passenger threshold of non-motor vehicles.

[0165] Among them, in step S5, the overload determination module can also be specifically implemented as the following technical solutions:

[0166] Directly process the partition analysis method of the image. The distribution of non-motor vehicles and people in the image is analyzed by partition, and the partition number is recorded as D zone (x, y).D zone (x, y) is represented by formula (8):

[0167] D zone (x, y) = ZoneID((x 1+x 2 ) / 2,(y 1 +y 2 ) / 2) (8);

[0168] Among them, D zone (x, y) is the partition number where the target (x, y) is located, and the partition matrix is ​​dynamically divided according to the image size.

[0169] The non-motor vehicle carrying passengers and overloading detection method based on road monitoring further includes step S6, which is located after step S5:

[0170] Step S6: To improve the overall performance of the detection, the image and video processing results are uniformly output in JSON format. The image and video processing results include the image ID k, the number of non-motor vehicles M, img (k) Average number of passengers And overload flag Overload(k).

[0171] Among them, the overload determination module can also be specifically implemented as the following technical solutions:

[0172] For the overload detection results of a single image, the cumulative overload probability P is predicted by the weighted probability model overload (k), P overload (k) is represented by formula (9):

[0173]

[0174] Among them, w ij is the weight of each target in the image, and δ(Overload(j)) is the overload status flag function.

[0175] Among them, the overload determination module can also be specifically implemented as the following technical solutions:

[0176] Combined with the characteristics of the partition, the statistical model of non-motor vehicles in the whole area is used to calculate the average passenger rate R of the partition. region , R region It is represented by formula (10):

[0177]

[0178] It is worth mentioning that the technical features such as the full-area non-motor vehicle statistical model involved in the patent application of this invention should be regarded as the prior art. The specific structure, working principle and possible control method and spatial layout method of these technical features can be selected according to the conventional selection in the field, and should not be regarded as the inventive point of the patent of this invention. The patent of this invention will not be further elaborated.

[0179] For those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A non-motor vehicle passenger and overload detection system based on road monitoring, characterized in that: include: A data acquisition module, which extracts pictures from the monitoring device or imports pictures from the monitoring device to the data acquisition module to output a first intermediate data set; An image preprocessing module, wherein the image preprocessing module acquires a first intermediate data set, and performs standardized image processing and contrast enhancement processing on the first intermediate data set to output a second intermediate data set; A target detection module, wherein the target detection module obtains a second intermediate data set, and the target detection module imports the second intermediate data set into a preset deep learning module to output non-motor vehicle and personnel targets; A people counting module, which obtains non-motor vehicles and human targets, and combines the non-motor vehicles and human targets with spatial analysis to output the number of people; The overload determination module obtains the number of personnel, determines whether the number of personnel is overloaded according to a preset threshold, and performs corresponding actions.

2. The non-motor vehicle passenger and overload detection system based on road monitoring according to claim 1 is characterized in that: The data acquisition module includes a video processing unit and an image processing unit, wherein: The video processing unit operates at a fixed frame rate F rate The proposed frame sequence {f i }; The image processing unit directly imports the image collection {I k }, the single picture number is k.

3. The non-motor vehicle passenger and overload detection system based on road monitoring according to claim 2 is characterized in that: The image preprocessing module includes a size normalization unit and an enhancement operation unit, wherein: Size normalization unit will be picture I k Adjust the size to S in , S in Characterized by formula (1): S in ＝640×640 (1)； The image after size standardization is denoted as I′ k , I′ k Characterized by formula (2): I′ k =Resize(I k ,S in ) (2); The enhancement operation unit optimizes the image quality by using a contrast enhancement algorithm (preferably using a CLAHE algorithm); The image after image quality optimization is denoted as I″ k , I″ k Characterized by formula (3): I″ k =CLAHE(I′ k ) (3)。 4. The non-motor vehicle passenger and overload detection system based on road monitoring according to claim 3 is characterized in that: The object detection module includes an object bounding box detection unit and a category label unit, where: The target bounding box detection unit processes the image I″ after image quality optimization k , to output the target bounding box b ij , target bounding box b ij Characterized by formula (4): b ij ={(x1,y1,x2,y2),c,s} (4); Among them, x1, y1 are the target bounding box b ij The upper left corner coordinates; x2, y2 are the target bounding box b ij The lower right corner coordinate of the target; c is the classification confidence of the target; s is the target bounding box b ij area; The category label unit processes the target bounding box b ij , to output non-motor vehicle category labels and personnel category labels; Single image target detection, use the YOLOv5 model or the YOLOv7 model to perform target detection to generate a set of bounding boxes B for each target img , the bounding box set B img It is represented by formula (5): B img ={b ij |j∈[1,N]} (5); Among them, B img is the target detection result set of the image, and N is the total number of detected targets.

5. The non-motor vehicle passenger and overload detection system based on road monitoring according to claim 4 is characterized in that: The people counting module is specifically implemented as the following technical solutions: Counting people in the image, by using the spatial relationship of the bounding box, counting the number of people N in each non-motor vehicle person (k), N person (k) is represented by formula (6): Among them, IoU(b ij , B motor ) is the overlap rate, and the value of δ(x) is as follows: when x>T iou When δ(x) is 1, otherwise δ(x) is 0; The overload determination module is specifically implemented as the following technical solutions: The overload determination method for images and videos combines the non-motor vehicle information detected in each image or video frame with the number of people. The statistically combined overload sign is recorded as Overload(k). Overload(k) is represented by formula (7): Where Overload(k) is the overload flag of picture k or video frame k, T max is the passenger threshold for non-motor vehicles; Directly process the partition analysis method of the image. The distribution of non-motor vehicles and people in the image is analyzed by partition, and the partition number is recorded as D zone (x,y),D zone (x, y) is represented by formula (8): D zone (x,y)=ZoneID((x1+x2) / 2,(y1+y2) / 2) (8); Among them, D zone (x, y) is the partition number where the target (x, y) is located, and the partition matrix is ​​dynamically divided according to the image size.

6. A method for detecting passengers and overloads on non-motor vehicles based on road monitoring, characterized in that: The following steps are involved: Step S1: the data acquisition module extracts pictures from the monitoring device or imports pictures from the monitoring device to the data acquisition module to output a first intermediate data set; Step S2: the image preprocessing module acquires a first intermediate data set, and performs standardized image processing and contrast enhancement processing on the first intermediate data set to output a second intermediate data set; Step S3: the target detection module acquires a second intermediate data set, and the target detection module imports the second intermediate data set into a preset deep learning module to output non-motor vehicle and human targets; Step S4: the people counting module obtains non-motor vehicles and people targets, and the people counting module combines the non-motor vehicles and people targets with the spatial analysis to output the number of people; Step S5: The overload determination module obtains the number of personnel, determines whether the number of personnel is overloaded according to a preset threshold, and performs corresponding actions.

7. The non-motor vehicle carrying passengers and overloading detection method based on road monitoring according to claim 6 is characterized in that: The data acquisition module includes a video processing unit and an image processing unit, wherein: The video processing unit operates at a fixed frame rate F rate The proposed frame sequence {f i }; The image processing unit directly imports the image collection {I k }, the single picture number is k.

8. The non-motor vehicle passenger and overload detection method based on road monitoring according to claim 7 is characterized in that: The image preprocessing module includes a size normalization unit and an enhancement operation unit, wherein: Size normalization unit will be picture I k Adjust the size to S in , S in Characterized by formula (1): S in ＝640×640 (1)； The image after size standardization is denoted as I′ k , I′ k Characterized by formula (2): I′ k =Resize(I k ,S in ) (2); The enhancement operation unit optimizes the image quality by using a contrast enhancement algorithm (preferably using a CLAHE algorithm); The image after image quality optimization is denoted as I″ k , I″ k Characterized by formula (3): I″ k =CLAHE(I′ k ) (3)。 9. The non-motor vehicle passenger and overload detection method based on road monitoring according to claim 8 is characterized in that: The object detection module includes an object bounding box detection unit and a category label unit, where: The target bounding box detection unit processes the image I″ after image quality optimization k , to output the target bounding box b ij , target bounding box b ij Characterized by formula (4): b ij ={(x1,y1,x2,y2),c,s} (4); Among them, x1, y1 are the target bounding box b ij The upper left corner coordinates; x2, y2 are the target bounding box b ij The lower right corner coordinate of the target; c is the classification confidence of the target; s is the target bounding box b ij area; The category label unit processes the target bounding box b ij , to output non-motor vehicle category labels and personnel category labels; Single image target detection, use the YOLOv5 model or the YOLOv7 model to perform target detection to generate a set of bounding boxes B for each target img , the bounding box set B img It is represented by formula (5): B img ={b ij |j∈[1,N]} (5); Among them, B img is the target detection result set of the image, and N is the total number of detected targets.

10. The non-motor vehicle passenger and overload detection method based on road monitoring according to claim 9, characterized in that: The people counting module is specifically implemented as the following technical solutions: Counting people in the image, by using the spatial relationship of the bounding box, counting the number of people N in each non-motor vehicle person (k), N person (k) is represented by formula (6): Among them, IoU(b ij , B motor ) is the overlap rate, and the value of δ(x) is as follows: when x>T iou When δ(x) is 1, otherwise δ(x) is 0; The overload determination module is specifically implemented as the following technical solutions: The overload determination method for images and videos combines the non-motor vehicle information detected in each image or video frame with the number of people. The statistically combined overload sign is recorded as Overload(k). Overload(k) is represented by formula (7): Where Overload(k) is the overload flag of picture k or video frame k, T max is the passenger threshold for non-motor vehicles; Directly process the partition analysis method of the image. The distribution of non-motor vehicles and people in the image is analyzed by partition, and the partition number is recorded as D zone (x,y),D zone (x, y) is represented by formula (8): D zone (x,y)=ZoneID((x1+x2) / 2,(y1+y2) / 2) (8); Among them, D zone (x, y) is the partition number where the target (x, y) is located, and the partition matrix is ​​dynamically divided according to the image size.

Images