A vehicle-mounted traffic pavement and accessory facility disease identification method

Abstract

The application provides a kind of vehicle-mounted traffic pavement and accessory facility disease identification method, comprising the following steps: cloud platform receives N video streams in real time, starts N threads;After video decoding to video stream, get picture frame;Picture frame is converted into two-dimensional array, and detection object detection and disease target detection are carried out in turn, obtain detection object GPS position information, detection object species information, disease category information and disease actual size information, and store and release.The vehicle-mounted traffic pavement and accessory facility disease identification method provided by the application can automatically distinguish and identify diseases using the video collected by the vehicle-mounted camera, output the latitude and longitude of the disease location, the size of the disease jurisdiction range, and the real-time push of dangerous and critical diseases, greatly reducing the inefficiency of manual identification, improving road maintenance efficiency, increasing safety during road operation, and having high practicality.

Description

Technical Field

[0001] This invention belongs to the field of defect identification technology, specifically relating to a method for identifying defects in vehicle-mounted road surfaces and ancillary facilities. Background Technology

[0002] Currently, the routine inspection method for highway defects is as follows: inspectors visually identify defects and manually record the results. This method has the following problems: it is easy to miss defects during highway defect inspections, and the efficiency of defect identification is relatively low, while the workload of inspectors is high. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for identifying defects in vehicle-mounted road surfaces and ancillary facilities, which can effectively solve the aforementioned problems.

[0004] The technical solution adopted in this invention is as follows:

[0005] This invention provides a method for identifying defects in vehicle-mounted road surfaces and ancillary facilities, comprising the following steps:

[0006] Step 1: N vehicle-mounted cameras are deployed in the area to be inspected; the video streams collected by each vehicle-mounted camera during the inspection are transmitted to the cloud platform in real time.

[0007] Step 2: The cloud platform receives N video streams in real time and starts N threads; where N video streams are represented as: video streams S1, S2, ..., S... N N threads are represented as: threads F1, F2, ..., F N ;

[0008] Each thread F i For video stream F i The analysis and processing are performed, where i = 1, 2, ..., N; the specific method is as follows:

[0009] Step 2.1, Thread F i For video stream F i Perform video decoding, converting the video stream F i Converted to a unified standard protocol format, the decoded video stream DF is obtained. i Among them, the decoded video stream DF i There are m image frames, denoted as: image frames tu1, tu2, ..., tu m ;

[0010] Step 2.2, process the decoded video stream DF i Preprocessing is performed, and each image frame is tu j Convert to a two-dimensional array to obtain image frames tu in two-dimensional array form. j [1];

[0011] Step 2.3: Employ the target detection model to process image frames in the form of a two-dimensional array. j [1] Perform object detection and determine the image frame tu in the form of a two-dimensional array. j [1] If there is no detected object, then directly delete the image frame tu in the form of a two-dimensional array. j [1] and end the process;

[0012] If so, then in the image frame tu in the form of a two-dimensional array. j In [1], the envelope rectangle of each detector is determined, and the envelope rectangle of each detector is cut out to obtain the detector slice of that detector;

[0013] For each detection object slice, denoted as: detection object slice P, the detection object slice P is desensitized by removing the background interference layer of the detection object slice P, and the feature map of the detection object slice after desensitization is obtained.

[0014] The desensitized analyte slice feature map is represented as: analyte slice feature map Feature(0). Then, proceed with steps 2.4-2.7:

[0015] Step 2.4: Obtain the basic information of the feature map (0) of the detected object slice, including: GPS location information of the detected object and type information of the detected object;

[0016] Add the basic information of the detection slice feature map Feature(0) to the detection slice feature map Feature(0) to obtain the detection slice feature map Feature(1);

[0017] Step 2.5: Perform disease target detection on Feature(1) of the detected material slice and determine whether there is a disease target in Feature(1). If not, the process ends.

[0018] If so, in the Feature(1) slice feature map of the detected object, the disease envelope rectangle of each disease is detected and drawn as the disease detection box of the disease; for any detected disease detection box, it is represented as: K-Feature(1); at the same time, the disease category information of the disease included in the disease detection box K-Feature(1) is identified;

[0019] For the disease detection frame K-Feature(1); search the disease standard database to obtain the size range of diseases of the same disease category;

[0020] If the size of the disease detection box K-Feature(1) is within the size range of diseases of the same disease category, then proceed to step 2.6; otherwise, the conclusion that the disease detection box K-Feature(1) is obviously unreasonable is obtained, and the disease detection box K-Feature(1) is discarded.

[0021] Step 2.6: Calculate the actual size of the disease in the disease detection frame K-Feature(1) to obtain the actual size information of the disease;

[0022] Step 3: Restore the feature map (Feature(1)) of the detected object slice containing the disease detection box K-Feature(1) to the image frame tu. j The disease detection box K-Feature(1) is marked, and the processed image frame tu is obtained. j [2]; At the same time, add the disease detection result information corresponding to the disease detection box K-Feature(1), including: GPS location information of the detected object, type information of the detected object, disease category information and actual size information of the disease;

[0023] Step 4: Save the disease detection results in text format and use data synchronization to store the corresponding processed image frames. j [2].

[0024] Preferably, step 2.6 specifically includes:

[0025] Step 2.6.1: Establish the projection relationship diagram between the vehicle-mounted camera and the defect detection frame K-Feature(1); wherein, the vehicle-mounted camera is the image frame tu corresponding to the defect detection frame K-Feature(1) that it has acquired. j Vehicle-mounted cameras;

[0026] In the projection diagram, the vehicle-mounted camera is represented as point A, and its image frame tu j When point A is projected onto the horizontal plane, point B is the projection point. Therefore, the installation height of the vehicle camera is the height L(AB), which is a known value.

[0027] The shooting angle range of the vehicle-mounted camera is ∠KAH. Specifically, with point A as the origin, the two boundary lines of its shooting range are: the upper boundary line AH and the lower boundary line AK. Point K is the intersection of the lower boundary line AK and the horizontal plane. The angle between the upper boundary line AH and the lower boundary line AK is ∠KAH, which is a known value.

[0028] Additionally, the angle between the lower boundary line AK and the vertical line AB can be obtained as ∠KAB;

[0029] Step 2.6.2, Calculation of the actual width of the disease detection frame K-Feature(1):

[0030] The initial end angle of the defect detection box K-Feature(1) captured by the vehicle camera is obtained. The meaning is: the scanning line of the vehicle camera starts from the lower boundary line AK and scans counterclockwise around point A. When it first scans any boundary point in the defect detection box K-Feature(1), assuming that the boundary point scanned is point D, then ∠DAK is the initial end angle.

[0031] The farthest angle of the defect detection box K-Feature(1) captured by the vehicle camera is obtained. The meaning is: the scanning line of the vehicle camera starts from the lower boundary line AK and scans counterclockwise around point A. Among the intersections of the vehicle camera and the defect detection box K-Feature(1), the boundary point farthest from point A is the farthest boundary point. Assuming that the farthest boundary point scanned is point E, then: ∠EAK is the farthest angle.

[0032] Therefore, the angle between the initial end and the farthest end is: ∠EAD=∠EAK-∠DAK;

[0033] Therefore, the actual width L(DE) of the disease detection frame K-Feature(1) is obtained by the following formula:

[0034] L(DE)=tan(∠EAD+∠DAK+∠KAB)*L(AB)–tan(∠DAK+∠KAB)*L(AB)

[0035] Step 2.6.3: Obtain the highest boundary point of the defect detection box K-Feature(1) captured by the vehicle camera. The meaning is: the scanning line of the vehicle camera starts from the lower boundary line AK and scans counterclockwise around point A. Among the intersections of the scanning line and the defect detection box K-Feature(1), the highest boundary point is the one furthest from the horizontal plane. Assume that the highest boundary point scanned is point F, and the intersection of the extension of line AF with the horizontal plane is point C.

[0036] Therefore, the included angle ∠CAE is obtained;

[0037] The distance L(BC) from point B to point C is calculated using the following formula:

[0038] L(BC)=tan(∠EAD+∠DAK+∠KAB+∠CAE)*L(AB)

[0039] The distance L(BE) from point B to point E is calculated using the following formula:

[0040] L(BE)=tan(∠EAD+∠DAK+∠KAB)*L(AB)

[0041] The actual height L(EF) of the disease detection frame K-Feature(1) is calculated using the following formula:

[0042] L(EF)=L(AB)*((L(BC)-L(BE)) / L(BC));

[0043] Thus, the actual width L(DE) and actual height L(EF) of the disease detection frame K-Feature(1) are obtained.

[0044] Preferred options also include:

[0045] Step 5, for video stream F i After analysis and processing, multiple processed image frames were obtained. j [2], multiple processed image frames tu j [2] Convert the disease into a short video and store it.

[0046] Preferred options also include:

[0047] Step 6: Transfer the disease detection results and processed image frames. j [2] And short videos of diseases are pushed to designated relevant personnel.

[0048] The method for identifying defects in vehicle-mounted road surfaces and ancillary facilities provided by this invention has the following advantages:

[0049] This invention provides a method for identifying road surface and ancillary facilities defects by using video captured by a vehicle-mounted camera. The method can automatically identify defects, output the latitude and longitude of the defect location, the size of the defect area, and real-time push notifications of dangerous and critical defects. This greatly reduces the inefficiency of manual identification, improves road maintenance efficiency, and increases safety during road operation and maintenance, making it highly practical. Attached Figure Description

[0050] Figure 1 A flowchart illustrating a method for identifying defects in vehicle-mounted traffic surfaces and ancillary facilities provided by the present invention;

[0051] Figure 2 This is a schematic diagram showing the actual dimensions of the disease detection frame provided by the present invention. Detailed Implementation

[0052] To make the technical problems solved, the technical solutions, and the beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the invention.

[0053] To address the issues of human error in detecting road defects and the inability to accurately assess their type and size during road inspections, this invention provides a method for identifying defects in vehicle-mounted road surfaces and ancillary facilities. This method is an AI-powered intelligent identification method for road inspections, which, compared to manual methods, can identify defect types and sizes faster, more conveniently, more efficiently, and more accurately, and push notifications to relevant personnel.

[0054] This invention is based on video analysis technology and utilizes neural network algorithms for intensive learning and analysis. The method primarily combines ffmpeg's video transcoding function to decode different video streams, then uses an object detection model for identification and calculation, and outputs the relevant calculation results. The method includes the following steps:

[0055] Capture video stream;

[0056] Use ffmpeg to decode video streams into a unified format that can be processed by a custom standard;

[0057] Then, the decoded video stream is preprocessed, and the image frames are sorted into a two-dimensional array;

[0058] A target detection model is used to perform real-time target detection on image frames in the form of a two-dimensional array, resulting in a target slice for each target. After desensitization processing, a target slice feature map is obtained. At the same time, the GPS location information and target type information of the target are obtained.

[0059] Disease target detection is performed on the feature map of the detected object slice to obtain the disease detection box for each disease, and at the same time, the disease category and the actual size of the disease are obtained;

[0060] The feature map of the detected object slice containing the disease detection box will be restored to the original image frame, and the disease tracking data will be identified, including: the GPS location information of the detected object, the type of detected object, the disease category and the actual size of the disease;

[0061] Disease tracking data, serving as disease detection results, is published along with corresponding image frames using a self-built Minos platform and pushed to a designated platform. The platform then issues real-time alerts or notifications based on the analyzed disease parameters. This invention can be deployed using Docker, with file storage on the Minos platform, offering simple and convenient deployment, flexible operation, and applicability to various scenarios such as highway pavement inspection.

[0062] refer to Figure 1 This invention provides a method for identifying defects in vehicle-mounted road surfaces and ancillary facilities. It can be a defect identification system for vehicle-mounted road surfaces and ancillary facilities, developed based on the DeepStream SDK and running on a GPU server, including the following steps:

[0063] Step 1: N vehicle-mounted cameras are deployed in the area to be inspected; the video streams collected by each vehicle-mounted camera during the inspection are transmitted to the cloud platform in real time.

[0064] Step 2: The cloud platform receives N video streams in real time and starts N threads; where N video streams are represented as: video streams S1, S2, ..., S... N N threads are represented as: threads F1, F2, ..., F N ;

[0065] In terms of specific implementation, N video streams require N ffmpeg threads, and need to adopt multiplexing, multi-threading, and asynchronous network I / O modes. It also needs to provide extremely low video transcoding latency, be able to convert video protocols H265 / H264 / AAC / G711 / OPUS into H264 single protocol input, and at the same time ensure the aspect ratio of subsequent AI recognition, so as to better calculate the size and area of ​​the lesions.

[0066] Each thread F i For video stream F i The analysis and processing are performed, where i = 1, 2, ..., N; the specific method is as follows:

[0067] Step 2.1, Thread F i For video stream F i Perform video decoding, converting the video stream F i Converted to a unified standard protocol format, the decoded video stream DF is obtained. i Among them, the decoded video stream DF i There are m image frames, denoted as: image frames tu1, tu2, ..., tu m ;

[0068] For example, it can be

[0069] Video streams obtained using different protocols such as RTSP / RTMP / HLS / HTTP-FLV / WebSocket-FLV / GB28181 / HTTP-TS / WebSocket-TS / HTTP-fMP4 / WebSocket-fMP4 / MP4 / WebRTC are decoded into a unified standard protocol format, such as H264, using ffmpeg.

[0070] Step 2.2, process the decoded video stream DF i Preprocessing is performed, and each image frame is tu j Convert to a two-dimensional array to obtain image frames tu in two-dimensional array form. j [1];

[0071] Step 2.3: Employ the target detection model to process image frames in the form of a two-dimensional array. j [1] Perform object detection and determine the image frame tu in the form of a two-dimensional array. j [1] If there is no detected object, then directly delete the image frame tu in the form of a two-dimensional array. j [1] and end the process; among them, the detected objects can be traffic surfaces and various auxiliary equipment, such as railings.

[0072] If so, then in the image frame tu in the form of a two-dimensional array. j In [1], the envelope rectangle of each detector is determined, and the envelope rectangle of each detector is cut out to obtain the detector slice of that detector;

[0073] For each detection object slice, denoted as: detection object slice P, the detection object slice P is desensitized by removing the background interference layer of the detection object slice P, and the feature map of the detection object slice after desensitization is obtained.

[0074] Therefore, by desensitizing the sample slice P, the background interference layer is removed, ensuring that the feature map after desensitization is free from the influence of abnormal monitoring range.

[0075] The desensitized analyte slice feature map is represented as: analyte slice feature map Feature(0). Then, proceed with steps 2.4-2.7:

[0076] This step involves slicing and desensitizing the image frames to ensure that the image frames can be taken within the required range, thereby reducing the impact of the boundaries of the captured image frames on AI recognition and thus reducing the error rate and false alarm rate.

[0077] Step 2.4: Obtain the basic information of the feature map (0) of the detected object slice, including: GPS location information of the detected object and type information of the detected object;

[0078] Add the basic information of the detection slice feature map Feature(0) to the detection slice feature map Feature(0) to obtain the detection slice feature map Feature(1);

[0079] In practical implementation, a custom data layer can be inserted into the Backbone and the final Head layer of the target detection model to add custom basic information, including the GPS location information and type of the detected object, providing slice feature maps of the detected object, and information integration. The specific method is as follows: mark the basic information to be added as a new head, and then mix it into the Backbone.

[0080] Step 2.5: Perform disease target detection on Feature(1) of the detected material slice and determine whether there is a disease target in Feature(1). If not, the process ends.

[0081] If so, in the Feature(1) slice feature map of the detected object, the disease envelope rectangle of each disease is detected and drawn as the disease detection box of the disease; for any detected disease detection box, it is represented as: K-Feature(1); at the same time, the disease category information of the disease included in the disease detection box K-Feature(1) is identified;

[0082] For the disease detection frame K-Feature(1); search the disease standard database to obtain the size range of diseases of the same disease category;

[0083] If the size of the disease detection box K-Feature(1) is within the size range of diseases of the same disease category, then proceed to step 2.6; otherwise, the conclusion that the disease detection box K-Feature(1) is obviously unreasonable is obtained, and the disease detection box K-Feature(1) is discarded.

[0084] Step 2.6: Calculate the actual size of the disease in the disease detection frame K-Feature(1) to obtain the actual size information of the disease;

[0085] The specific method is as follows: gather information such as the camera's shooting position, use OpenCV to draw the detection box for the disease, and use the image normalization algorithm to calculate the actual size of the disease detection box.

[0086] refer to Figure 2 Step 2.6 specifically involves:

[0087] Step 2.6.1: Establish the projection relationship diagram between the vehicle-mounted camera and the defect detection frame K-Feature(1); wherein, the vehicle-mounted camera is the image frame tu corresponding to the defect detection frame K-Feature(1) that it has acquired. j Vehicle-mounted cameras;

[0088] In the projection diagram, the vehicle-mounted camera is represented as point A, and its image frame tu j When point A is projected onto the horizontal plane, point B is the projection point. Therefore, the installation height of the vehicle camera is the height L(AB), which is a known value.

[0089] The shooting angle range of the vehicle-mounted camera is ∠KAH. Specifically, with point A as the origin, the two boundary lines of its shooting range are: the upper boundary line AH and the lower boundary line AK. Point K is the intersection of the lower boundary line AK and the horizontal plane. The angle between the upper boundary line AH and the lower boundary line AK is ∠KAH, which is a known value.

[0090] Additionally, the angle between the lower boundary line AK and the vertical line AB can be obtained as ∠KAB;

[0091] Step 2.6.2, Calculation of the actual width of the disease detection frame K-Feature(1):

[0092] The initial end angle of the defect detection box K-Feature(1) captured by the vehicle camera is obtained. The meaning is: the scanning line of the vehicle camera starts from the lower boundary line AK and scans counterclockwise around point A. When it first scans any boundary point in the defect detection box K-Feature(1), assuming that the boundary point scanned is point D, then ∠DAK is the initial end angle.

[0093] The farthest angle of the defect detection box K-Feature(1) captured by the vehicle camera is obtained. The meaning is: the scanning line of the vehicle camera starts from the lower boundary line AK and scans counterclockwise around point A. Among the intersections of the vehicle camera and the defect detection box K-Feature(1), the boundary point farthest from point A is the farthest boundary point. Assuming that the farthest boundary point scanned is point E, then: ∠EAK is the farthest angle.

[0094] Therefore, the angle between the initial end and the farthest end is: ∠EAD=∠EAK-∠DAK;

[0095] Therefore, the actual width L(DE) of the disease detection frame K-Feature(1) is obtained by the following formula:

[0096] L(DE)=tan(∠EAD+∠DAK+∠KAB)*L(AB)–tan(∠DAK+∠KAB)*L(AB)

[0097] Step 2.6.3: Obtain the highest boundary point of the defect detection box K-Feature(1) captured by the vehicle camera. The meaning is: the scanning line of the vehicle camera starts from the lower boundary line AK and scans counterclockwise around point A. Among the intersections of the scanning line and the defect detection box K-Feature(1), the highest boundary point is the one furthest from the horizontal plane. Assume that the highest boundary point scanned is point F, and the intersection of the extension of line AF with the horizontal plane is point C.

[0098] Therefore, the included angle ∠CAE is obtained;

[0099] The distance L(BC) from point B to point C is calculated using the following formula:

[0100] L(BC)=tan(∠EAD+∠DAK+∠KAB+∠CAE)*L(AB)

[0101] The distance L(BE) from point B to point E is calculated using the following formula:

[0102] L(BE)=tan(∠EAD+∠DAK+∠KAB)*L(AB)

[0103] The actual height L(EF) of the disease detection frame K-Feature(1) is calculated using the following formula:

[0104] L(EF)=L(AB)*((L(BC)-L(BE)) / L(BC));

[0105] Thus, the actual width L(DE) and actual height L(EF) of the disease detection frame K-Feature(1) are obtained.

[0106] Step 3: Restore the feature map (Feature(1)) of the detected object slice containing the disease detection box K-Feature(1) to the image frame tu. j The disease detection box K-Feature(1) is marked, and the processed image frame tu is obtained. j [2]; At the same time, add the disease detection result information corresponding to the disease detection box K-Feature(1), including: GPS location information of the detected object, type information of the detected object, disease category information and actual size information of the disease;

[0107] Step 4: Save the disease detection results in text format and use data synchronization to store the corresponding processed image frames. j [2].

[0108] Also includes:

[0109] Step 5, for video stream F i After analysis and processing, multiple processed image frames were obtained. j [2], multiple processed image frames tu j [2] Convert the disease into a short video and store it.

[0110] Also includes:

[0111] Step 6: Transfer the disease detection results and processed image frames. j [2] And short videos of diseases are pushed to designated relevant personnel.

[0112] In practical implementation, after obtaining the disease detection results, this information can be sent to the backend data storage as disease data elements. Simultaneously, the Minos data synchronization function is invoked to store relevant disease images and short videos as files. Combined with the platform's alarm system, disease data and alarms are pushed out as required, promptly notifying relevant personnel of the disease situation. For example, images and converted short videos are saved to Minos storage. If a particularly dangerous or critical disease occurs, the system will push a message to designated personnel.

[0113] This invention uses a vehicle-mounted camera and a fixed-point PTZ monitoring system equipped with AI video algorithms to automatically identify and filter out defects in the captured videos and photos. Furthermore, it utilizes projection algorithms to automatically calculate relevant data on defects, thereby reducing missed detections and obtaining more accurate data on road surface and ancillary facilities defects.

[0114] Therefore, the present invention provides a method for identifying defects in vehicle-mounted road surfaces and ancillary facilities. By utilizing video collected by a vehicle-mounted camera, it can automatically identify defects, output the latitude and longitude of the defect location, the size of the defect's area, and real-time push notifications of dangerous and critical defects. This greatly reduces the inefficiency of manual identification, improves road maintenance efficiency, and increases safety during road operation and maintenance, making it highly practical.

[0115] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying diseases of a vehicle-mounted traffic pavement and an accessory, characterized by, The method comprises the following steps: Step 1, arranging N vehicle-mounted cameras in the patrolled area; The video stream collected by each vehicle-mounted camera during the patrol is transmitted to the cloud platform in real time; Step 2, the cloud platform receives N video streams in real time, and starts N threads; wherein, the N video streams are represented as: video streams S1, S2, …, S N N threads are represented as: threads F1, F2, …, F N ​ Each thread F i analyzes the video stream F i , where i = 1, 2, …, N; the specific manner is: Step 2.1, thread F i decoding the video stream F i into a format of a unified standard protocol to obtain a decoded video stream DF i i ; wherein the decoded video stream DF i is m picture frames, respectively represented as: picture frames tu1, tu2, …, tu m ;​ Step 2.2, pre-process the decoded video stream DF i to convert each picture frame tu j into a two-dimensional array, obtaining a picture frame tu j [1] in the form of a two-dimensional array; Step 2.3, using the detection object detection model to detect the picture frame tu in the form of a two-dimensional array j [1] to determine whether the picture frame tu in the form of a two-dimensional array has a detection object j [1], if not, directly delete the picture frame tu in the form of a two-dimensional array j [1], and end the process; If yes, the picture frame tu in the form of a two-dimensional array j [1] The envelope rectangular frame of each detection object is determined, and the envelope rectangular frame of each detection object is cut down to obtain a detection object slice of the detection object. For each detection slice, denoted as detection slice P, the detection slice P is desensitized, and the background interference layer of the detection slice P is removed to obtain a detection slice feature map after desensitization; For the detection slice feature map after desensitization, denoted as detection slice feature map Feature(0), steps 2.4-2.7 are executed: Step 2.4, obtaining the basic information of the detection slice feature map Feature(0), including the detection object GPS position information and the detection object category information; The basic information of the detection slice feature map Feature(0) is added to the detection slice feature map Feature(0) to obtain a detection slice feature map Feature(1); Step 2.5, performing disease target detection on the detection slice feature map Feature(1) to determine whether there is a disease target in the detection slice feature map Feature(1); if not, the process is ended; If there is, the disease envelope rectangular frame of each disease in the detection slice feature map Feature(1) is detected and drawn as a disease detection frame; for any disease detection frame, denoted as K-Feature(1); at the same time, the disease category information of the disease included in the disease detection frame K-Feature(1) is identified; For the disease detection frame K-Feature(1), the disease size range of the same disease category is obtained by searching the disease standard database; If the size of the disease detection frame K-Feature(1) is located in the disease size range of the same disease category, step 2.6 is executed; otherwise, the conclusion that the disease detection frame K-Feature(1) is obviously unreasonable is obtained, and the disease detection frame K-Feature(1) is discarded; Step 2.6, calculating the actual size of the disease detection frame K-Feature(1) to obtain the actual size information of the disease; Step 3, reduce the detection object slice feature map Feature (1) containing the disease detection frame K-Feature (1) to the picture frame tu j And identify the disease detection frame K-Feature (1), get the processed picture frame tu j [2]; At the same time, add the disease detection result information corresponding to the disease detection frame K-Feature (1), including: detection object GPS position information, detection object species information, disease category information and disease actual size information; Step 4, save the disease detection result information in text form, and store the corresponding processed picture frame tu by using the data synchronization function j [2] 2. The method of claim 1, wherein the method is a method of identifying a disease of a road surface and an accessory facility for a vehicle. Step 2.6 is specifically: Step 2.6.1, establish the projection relationship diagram of the vehicle-mounted camera and the disease detection frame K-Feature(1); wherein the vehicle-mounted camera is the vehicle-mounted camera that collects the picture frame tu corresponding to the disease detection frame K-Feature(1) j the vehicle-mounted camera; In the projection relationship diagram, the vehicle-mounted camera is represented as point A, which collects a picture frame tu j When point A is in the horizontal plane, the projection point of point A is point B, and thus the installation height of the vehicle-mounted camera, i.e., the height L(AB), is a known value; The shooting angle range of the vehicle-mounted camera is ∠KAH, specifically, taking point A as the origin, the two boundary lines of the shooting range are upper boundary line AH and lower boundary line AK; wherein, point K is the intersection of lower boundary line AK and horizontal plane; the included angle between upper boundary line AH and lower boundary line AK is ∠KAH, which is a known value; In addition, the included angle between lower boundary line AK and vertical line AB is ∠KAB; Step 2.6.2, calculation of the actual width of the disease detection frame K-Feature(1): The initial end angle of the disease detection frame K-Feature(1) shot by the vehicle-mounted camera is obtained, which means that the scanning line of the vehicle-mounted camera starts from the lower boundary line AK and scans counterclockwise around point A, and first scans to any boundary point in the disease detection frame K-Feature(1); assuming that the scanned boundary point is point D, then ∠DAK is the initial end angle; Obtaining the farthest end angle of the disease detection frame K-Feature(1) shot by the vehicle-mounted camera, which means that the scanning line of the vehicle-mounted camera scans counterclockwise from the lower boundary line AK around point A, and in the intersection of the disease detection frame K-Feature(1), the boundary point farthest from point A. Assuming that the farthest boundary point scanned is point E, then ∠EAK is the farthest end angle; Thus, the included angle from the initial end to the farthest end is ∠EAD=∠EAK-∠DAK; Thus, the actual width L(DE) of the disease detection frame K-Feature(1) is obtained by the following formula: L(DE)=tan(∠EAD+∠DAK+∠KAB)*L(AB)-tan(∠DAK+∠KAB)*L(AB) Step 2.6.3, obtaining the highest boundary point of the disease detection frame K-Feature(1) shot by the vehicle-mounted camera, which means that the scanning line of the vehicle-mounted camera scans counterclockwise from the lower boundary line AK around point A, and in the intersection of the disease detection frame K-Feature(1), the boundary point farthest from the horizontal plane. Assuming that the highest boundary point scanned is point F, and the intersection of the extension line of line AF and the horizontal plane is point C; Thus, the included angle ∠CAE is obtained; The distance L(BC) from point B to point C is calculated by the following formula: L(BC)=tan(∠EAD+∠DAK+∠KAB+∠CAE)*L(AB) The distance L(BE) from point B to point E is calculated by the following formula: L(BE)=tan(∠EAD+∠DAK+∠KAB)*L(AB) The actual height L(EF) of the disease detection frame K-Feature(1) is calculated by the following formula: L(EF)=L(AB)*((L(BC)-L(BE)) / L(BC)); Thus, the actual width L(DE) and the actual height L(EF) of the disease detection frame K-Feature(1) are obtained.

3. The method of claim 1, wherein the method is a method of identifying a disease of a road surface and an accessory facility of a vehicle, characterized by, Also including: Step 5, for the video stream F i After analysis and processing, a plurality of processed picture frames tu j [2] are obtained j [2] are converted into short videos of diseases for storage.

4. The method of claim 3, wherein the method further comprises: Also including: Step 6, the disease detection result information, the processed picture frame tu j [2] and disease short video to the designated relevant personnel.

Images