Monocular depth ranging recognition-based berth shielding judgment method and system
Through the combination of monocular depth ranging and object detection model, the accuracy of berth occlusion judgment in video pile scenarios is solved, and more efficient berth parking fee management is achieved.
Patent Information
- Application Number
- CN202411716088.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-27
- Publication Date
- 2025-03-18
AI Technical Summary
It is difficult for the prior art to accurately determine whether the berth is blocked in the video pile scene, especially when the shape, color and size of the vehicle or foreign object vary in various changes, resulting in false alarms or missed alarms.
Through monocular depth ranging combined with vehicle coordinates and chassis information, the vehicle detection model is used to perform vehicle detection, and the occlusion is judged through depth information and pixel ratio, the vehicle and the occlusion are separated, and then whether the berth is blocked.
The accuracy of judgment that berths or vehicles are blocked in video pile scenarios is improved, thereby improving the charging efficiency of on-road berth parking.
Smart Images
Figure CN119649318A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of pattern recognition, and in particular to a berth obstruction judgment method based on monocular depth ranging recognition and a berth obstruction judgment system based on monocular depth ranging recognition. Background Art
[0002] In on-street parking projects, due to the low installation height of the video pole camera, the vehicles in the parking space are easily blocked by other objects, resulting in the camera being unable to correctly identify the vehicle license plate number, or the parking space is occupied by other foreign objects, making it impossible to park the vehicle normally, ultimately causing the operation and management party to lose the transaction order; some car owners even take advantage of related loopholes and use obstructions to block the camera, thereby evading paying parking fees.
[0003] At present, there are mainly the following ways to implement the existing occlusion judgment algorithm:
[0004] 1. Judging whether a parking space is blocked by judging whether there are large vehicles parked in adjacent parking spaces is relatively rough and cannot adapt to the video pile scene; detecting foreign objects in parking spaces by target detection also has certain limitations. For example, the types, shapes, colors or sizes of foreign objects vary greatly, and the detection model cannot list them all.
[0005] 2. Simply calculate the intersection and union ratio of the parked vehicle and the parking area to be greater than a certain threshold to determine whether the parking space is blocked. The premise is that the vehicle must be detected; and the empty space detection is also based on the target detection model. Its disadvantages are the same as the parking space foreign object detection in "Parking Space Status Perception and Judgment Method, Device, Computer Equipment and Storage Medium".
[0006] 3. Select 6 key points of the berth for detection. If the first key point, the second key point, the third key point and the sixth key point are detected, it is determined that the berth to be detected is not blocked. Otherwise, it is determined that the berth to be detected is blocked. This method has poor adaptability. When the berth line is not clear, it is easy to cause key points to be missed, resulting in false alarms of occlusion; foreign objects on the road surface may cause multiple key points to be detected, resulting in false alarms of occlusion.
[0007] 4. The side image of the vehicle is collected through a monocular camera, and a depth ranging model is constructed through a stereo geometric imaging model, a distortion correction model, calibration coefficients, detection verification and stereo matching. The monocular depth ranging of the pedestrian / vehicle on the side is obtained through calculation. However, its technical principle is essentially unable to achieve the effect of the current deep learning ranging model, and is only suitable for a certain distance and when the texture of the object is clear.
[0008] 5. Use a monocular depth estimation network model to obtain the depth information of the image objects, and segment the depth information into connected domains. If the maximum average depth area in the image is greater than a certain threshold, the camera is considered to be blocked; otherwise, the image blur is judged, and if it is greater than a certain threshold, the camera is also considered to be blocked. This method is prone to misjudging the road surface as an obstruction, because in the video pile scene, the camera is installed low, the road surface area is close to the camera, and the road surface area accounts for a large proportion of the image. It only supports the judgment of larger obstructions, and it is impossible to clearly determine whether the vehicle is blocked or the parking space is blocked by foreign objects. Summary of the invention
[0009] In view of the above problems, the present invention provides a parking space obstruction judgment method and system based on monocular depth ranging recognition. Through monocular depth ranging and combined with vehicle coordinates and vehicle chassis information, a target detection model is used to detect vehicles, and depth information and pixel ratio are used to judge obstructions. It can effectively judge whether the vehicle in the on-street parking space in the video pole scene is obstructed, improve the accuracy of judgment on whether the parking space or vehicle is obstructed, and thus improve the charging efficiency of on-street parking spaces.
[0010] To achieve the above object, the present invention provides a berth occlusion judgment method based on monocular depth ranging recognition, comprising:
[0011] Acquire an image of the target berth area, and extract depth ranging information from the image;
[0012] Performing vehicle detection on the image based on a preset target detection model;
[0013] If a vehicle is detected, determining whether the vehicle is in the target parking space according to the intersection of the chassis quadrilateral of the vehicle and the quadrilateral of the target parking space;
[0014] If the vehicle is in the target berth, separating the vehicle from the obstruction according to the depth ranging information;
[0015] Determining whether the vehicle is blocked according to a pixel ratio between the separated blocking object and the vehicle;
[0016] If no vehicle is detected, the edge of the obstruction is extracted for the target parking space, and whether the target parking space is obstructed is determined according to the pixel ratio of the obstruction to the target parking space.
[0017] In the above implementation, preferably, the process of acquiring an image of the target berth area and extracting depth ranging information in the image includes:
[0018] Using video poles arranged near the target berth, an image corresponding to the target berth area is collected;
[0019] Performing depth estimation on the image using a depth model trained with a data set;
[0020] Normalize the obtained depth distance information and convert it into a visual depth map;
[0021] From left to right and from bottom to top, calculate the difference between each pixel in the depth map and the previous N and previous N*2 pixels, and set the current pixel to 0 when the difference with the previous N*2 pixels is greater than the difference with the previous N pixels, and both differences are greater than a preset threshold, until the entire depth map is traversed.
[0022] In the above implementation, preferably, the process of judging whether the vehicle is in the target berth according to the intersection of the chassis quadrilateral of the vehicle and the quadrilateral of the target berth includes:
[0023] Calculating the intersection area of the vehicle's chassis quadrilateral and the target berth quadrilateral;
[0024] If the intersection area is greater than a preset threshold, it is determined that the vehicle is in the target parking space.
[0025] In the above implementation, preferably, the process of separating the vehicle from the obstruction according to the depth ranging information and judging whether the vehicle is obstructed according to the pixel ratio of the separated obstruction and the vehicle includes:
[0026] By utilizing the different depth information of the front and rear objects, when a vehicle and an obstruction exist in the image, the vehicle is separated from the obstruction according to the depth ranging information;
[0027] The pixel area ratio of the obstruction and the vehicle is calculated, and when the pixel area ratio is greater than a preset threshold, it is determined that the vehicle is obstructed, otherwise it is determined that the vehicle is not obstructed.
[0028] In the above implementation, preferably, the step of extracting the edge of the occluder on the target berth and judging whether the target berth is blocked according to the pixel ratio between the occluder and the target berth includes:
[0029] When there is an obstruction in the target berth area, extract edge information of the obstruction in the target berth according to the depth ranging information, and use the erosion morphology algorithm to denoise the edge information;
[0030] Fitting the edge of the occluder into a closed figure through contour search and a distance-based closing algorithm;
[0031] The ratio of the pixel area of the closed figure to the pixel area of the target berth is calculated, and when the ratio is greater than a preset threshold, it is determined that the target berth is blocked, otherwise it is determined that the target berth is not blocked.
[0032] In the above implementation, preferably, the berth occlusion judgment method based on monocular depth ranging recognition also includes:
[0033] If the target berth is not blocked, clustering calculation is performed on the depth map using a clustering algorithm, and objects in the map are classified into three categories: far, medium, and near according to the depth information;
[0034] Calculating a connected domain of a nearby object, and calculating average depth information of the nearby object based on the connected domain;
[0035] If the average depth of the nearby object is greater than a specified threshold, and the ratio of the object area to the image area is greater than a preset threshold, it is determined that the image is occluded; otherwise, it is determined that the image is not occluded.
[0036] The present invention further proposes a berth occlusion judgment system based on monocular depth ranging recognition, which uses the berth occlusion judgment method based on monocular depth ranging recognition disclosed in any one of the above embodiments, including:
[0037] A depth image acquisition module, used to acquire an image of a target berth area and extract depth ranging information from the image;
[0038] A vehicle model detection module, used to perform vehicle detection on the image based on a preset target detection model;
[0039] A vehicle presence determination module, for determining whether the vehicle is in the target berth according to the intersection of the chassis quadrilateral of the vehicle and the quadrilateral of the target berth when a vehicle is detected;
[0040] A vehicle shielding separation module, used for separating the vehicle from the shielding object according to the depth ranging information when the vehicle is located in the target berth;
[0041] A vehicle occlusion judgment module, used to judge whether the vehicle is occluded according to the pixel ratio of the separated occluding object and the vehicle;
[0042] The parking space blocking judgment module is used to extract the edge of the blocking object for the target parking space when no vehicle is detected, and judge whether the target parking space is blocked according to the pixel ratio of the blocking object to the target parking space.
[0043] In the above implementation, preferably, the berth occlusion judgment system based on monocular depth ranging recognition further includes an image occlusion judgment module, which is specifically used to:
[0044] When the target berth is not blocked, clustering calculation is performed on the depth map using a clustering algorithm, and objects in the map are classified into three categories: far, medium, and near according to depth information;
[0045] Calculating a connected domain of a nearby object, and calculating average depth information of the nearby object based on the connected domain;
[0046] If the average depth of the nearby object is greater than a specified threshold, and the ratio of the object area to the image area is greater than a preset threshold, it is determined that the image is occluded; otherwise, it is determined that the image is not occluded.
[0047] In the above implementation, preferably, the vehicle shielding separation module is specifically used for:
[0048] By utilizing the different depth information of the front and rear objects, when a vehicle and an obstruction exist in the image, the vehicle is separated from the obstruction according to the depth ranging information;
[0049] The pixel area ratio of the obstruction and the vehicle is calculated, and when the pixel area ratio is greater than a preset threshold, it is determined that the vehicle is obstructed, otherwise it is determined that the vehicle is not obstructed.
[0050] In the above implementation, preferably, the vehicle blocking judgment module is specifically used for:
[0051] When there is an obstruction in the target berth area, extract edge information of the obstruction in the target berth according to the depth ranging information, and use the erosion morphology algorithm to denoise the edge information;
[0052] Fitting the edge of the occluder into a closed figure through contour search and a distance-based closing algorithm;
[0053] The ratio of the pixel area of the closed figure to the pixel area of the target berth is calculated, and when the ratio is greater than a preset threshold, it is determined that the target berth is blocked, otherwise it is determined that the target berth is not blocked.
[0054] Compared with the prior art, the beneficial effects of the present invention are as follows: through monocular depth ranging combined with vehicle coordinates and vehicle chassis information, the target detection model is used to detect the vehicle, and the depth information and pixel ratio are used to judge the obstruction. It is possible to effectively judge whether the vehicle in the on-street parking space in the video pole scene is obstructed, thereby improving the accuracy of judging whether the parking space or the vehicle is obstructed, thereby improving the charging efficiency of on-street parking spaces. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 A schematic flow chart of a method for determining berth occlusion based on monocular depth ranging recognition disclosed in an embodiment of the present invention;
[0056] Figure 2 The present invention is a schematic diagram of a module of a berth occlusion judgment system based on monocular depth ranging recognition disclosed in an embodiment of the present invention.
[0057] In the figure, the corresponding relationship between each component and the reference numeral is as follows:
[0058] 1. Depth image acquisition module, 2. Vehicle model detection module, 3. Vehicle presence judgment module, 4. Vehicle occlusion separation module, 5. Vehicle occlusion judgment module, 6. Berth occlusion judgment module, 7. Image occlusion judgment module. DETAILED DESCRIPTION
[0059] In order to make the purpose, implementation mode and advantages of the embodiments of the present invention clearer, the implementation mode in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0060] The present invention is further described in detail below in conjunction with the accompanying drawings:
[0061] like Figure 1 As shown, a berth occlusion judgment method based on monocular depth ranging recognition provided by the present invention includes:
[0062] Acquire an image of the target berth area and extract depth ranging information from the image;
[0063] Perform vehicle detection on the image based on the preset target detection model;
[0064] If a vehicle is detected, it is determined whether the vehicle is in the target parking space based on the intersection of the vehicle's chassis quadrilateral and the target parking space quadrilateral;
[0065] If the vehicle is in the target parking space, the vehicle is separated from the obstruction based on the depth ranging information;
[0066] Determine whether the vehicle is blocked based on the pixel ratio of the separated occluder and the vehicle;
[0067] If no vehicle is detected, the edge of the occluder is extracted for the target parking space, and whether the target parking space is blocked is determined based on the pixel ratio between the occluder and the target parking space.
[0068] In this implementation, monocular depth ranging is used in combination with vehicle coordinates and vehicle chassis information, a target detection model is used for vehicle detection, and depth information and pixel ratio are used for occlusion judgment. This can effectively determine whether a vehicle in an on-street parking space is obstructed in a video pole scene, thereby improving the accuracy of determining whether a parking space or vehicle is obstructed, thereby improving the charging efficiency of on-street parking spaces.
[0069] In the above implementation, preferably, an image of the target berth area is acquired, and depth ranging information in the image is extracted. The specific process includes:
[0070] Using the video poles pre-set near the target berth, collect images corresponding to the target berth area;
[0071] The depth model trained with DA-2K, Hypersim and vKITTI datasets is used to estimate the depth of the image.
[0072] Normalize the obtained depth distance information and convert it into a visible depth map; for example, normalize the pixel values of objects from far to near to 0 to 255 respectively. The farther the distance, the smaller the pixel value. Assuming that the minimum and maximum depth values are min and max respectively, and the current pixel depth value is val, the corresponding grayscale value pix is calculated as follows:
[0073] pix=(val-min) / (max-min)*255
[0074] Select the column data of the depth map from left to right, and calculate the difference V1 and V2 between each pixel in the depth map and the previous N and N*2 pixels in order from bottom to top. When V2>V1 and V1 and V2 are both greater than the preset threshold, set the current pixel to 0 until the entire depth map is traversed.
[0075] In the above embodiment, preferably, judging whether the vehicle is in the target berth is based on the intersection of the chassis quadrilateral of the vehicle and the quadrilateral of the target berth. The specific process includes:
[0076] Calculate the intersection area between the vehicle's chassis quadrilateral and the target parking space quadrilateral;
[0077] If the intersection area is greater than a preset threshold, the vehicle is determined to be in the target parking space.
[0078] In the above implementation, preferably, the vehicle and the obstruction are separated according to the depth ranging information, and whether the vehicle is obstructed is determined according to the pixel ratio of the separated obstruction and the vehicle. The specific process includes:
[0079] By taking advantage of the different depth information of the front and rear objects, the vehicle and the occlusion are separated when there are vehicles and occlusions in the image according to the depth ranging information.
[0080] The number of pixels where the separated occluders fall on the vehicle is counted, and the pixel area ratio of the occluder to the vehicle is calculated. When the pixel area ratio is greater than a preset threshold, the vehicle is determined to be occluded, otherwise the vehicle is determined to be not occluded.
[0081] In the above implementation, preferably, the edge of the occluder is extracted for the target berth, and whether the target berth is occluded is determined according to the pixel ratio of the occluder to the target berth. The specific process includes:
[0082] When there are obstructions in the target berth area, the Canny edge information of the obstructions in the target berth is extracted according to the depth ranging information, and the erosion morphology algorithm is used to denoise the edge information;
[0083] Through contour search and distance-based closing algorithm, the edge of the occluder is fitted into a closed shape;
[0084] The ratio of the pixel area of the closed figure to the pixel area of the target berth is calculated, and when the ratio is greater than a preset threshold, it is determined that the target berth is blocked, otherwise it is determined that the target berth is not blocked.
[0085] In the above implementation, preferably, the berth occlusion judgment method based on monocular depth ranging recognition also includes:
[0086] If the target berth is not blocked, the K-Means clustering algorithm is used to perform clustering calculations on the depth map, and objects in the map are classified into three categories: far, medium, and near according to the depth information;
[0087] Calculate the connected domain of nearby objects, and calculate the average depth information of nearby objects based on the connected domain;
[0088] If the average depth of the nearby object is greater than the specified threshold and the ratio of the object area to the image area is greater than the preset threshold, it is determined that the image is occluded; otherwise, it is determined that the image is not occluded.
[0089] like Figure 2 As shown, the present invention further proposes a berth occlusion judgment system based on monocular depth ranging recognition, which applies the berth occlusion judgment method based on monocular depth ranging recognition disclosed in any one of the above embodiments, including:
[0090] The depth image acquisition module 1 is used to acquire an image of the target berth area and extract the depth ranging information in the image;
[0091] The vehicle model detection module 2 is used to detect vehicles in the image based on a preset target detection model;
[0092] The vehicle presence determination module 3 is used to determine whether the vehicle is in the target parking space according to the intersection of the vehicle chassis quadrilateral and the target parking space quadrilateral when the vehicle is detected;
[0093] The vehicle blocking separation module 4 is used to separate the vehicle from the blocking object according to the depth ranging information when the vehicle is located in the target berth;
[0094] The vehicle occlusion judgment module 5 is used to judge whether the vehicle is occluded according to the pixel ratio of the separated occluder and the vehicle;
[0095] The parking space blocking judgment module 6 is used to extract the edge of the blocking object for the target parking space when no vehicle is detected, and judge whether the target parking space is blocked according to the pixel ratio between the blocking object and the target parking space.
[0096] In the above implementation, preferably, the berth occlusion judgment system based on monocular depth ranging recognition further includes an image occlusion judgment module 7, which is specifically used to:
[0097] When the target berth is not blocked, the clustering algorithm is used to perform clustering calculation on the depth map, and the objects in the map are classified into three categories: far, medium, and near according to the depth information;
[0098] Calculate the connected domain of nearby objects, and calculate the average depth information of nearby objects based on the connected domain;
[0099] If the average depth of the nearby object is greater than the specified threshold and the ratio of the object area to the image area is greater than the preset threshold, it is determined that the image is occluded; otherwise, it is determined that the image is not occluded.
[0100] In the above embodiment, preferably, the vehicle shielding separation module 4 is specifically used for:
[0101] By taking advantage of the different depth information of the front and rear objects, the vehicle and the occlusion are separated when there are vehicles and occlusions in the image according to the depth ranging information.
[0102] The pixel area ratio of the obstruction and the vehicle is calculated. When the pixel area ratio is greater than a preset threshold, the vehicle is determined to be obstructed, otherwise the vehicle is determined to be not obstructed.
[0103] In the above implementation, preferably, the vehicle blocking judgment module 5 is specifically used for:
[0104] When there are obstructions in the target berth area, the edge information of the obstructions in the target berth is extracted according to the depth ranging information, and the erosion morphology algorithm is used to denoise the edge information;
[0105] Through contour search and distance-based closing algorithm, the edge of the occluder is fitted into a closed shape;
[0106] The ratio of the pixel area of the closed figure to the pixel area of the target berth is calculated, and when the ratio is greater than a preset threshold, it is determined that the target berth is blocked, otherwise it is determined that the target berth is not blocked.
[0107] According to the berth occlusion judgment system based on monocular depth ranging recognition disclosed in the above-mentioned embodiment, the functions to be implemented by each module thereof correspond to the steps of the berth occlusion judgment method based on monocular depth ranging recognition disclosed in the above-mentioned embodiment. During the implementation process, the operation is performed with reference to the above-mentioned embodiment, which will not be repeated here.
[0108] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A berth occlusion judgment method based on monocular depth ranging recognition, characterized in that: include: Acquire an image of the target berth area, and extract depth ranging information from the image; Performing vehicle detection on the image based on a preset target detection model; If a vehicle is detected, determining whether the vehicle is in the target parking space according to the intersection of the chassis quadrilateral of the vehicle and the quadrilateral of the target parking space; If the vehicle is in the target berth, separating the vehicle from the obstruction according to the depth ranging information; Determining whether the vehicle is blocked according to a pixel ratio between the separated blocking object and the vehicle; If no vehicle is detected, the edge of the obstruction is extracted for the target parking space, and whether the target parking space is obstructed is determined according to the pixel ratio of the obstruction to the target parking space.
2. The method for determining berth obstruction based on monocular depth ranging recognition according to claim 1, characterized in that: The process of acquiring an image of the target berth area and extracting depth ranging information from the image specifically includes: Using video poles arranged near the target berth, an image corresponding to the target berth area is collected; Performing depth estimation on the image using a depth model trained with a data set; Normalize the obtained depth distance information and convert it into a visual depth map; From left to right and from bottom to top, calculate the difference between each pixel in the depth map and the previous N and previous N*2 pixels, and set the current pixel to 0 when the difference with the previous N*2 pixels is greater than the difference with the previous N pixels, and both differences are greater than a preset threshold, until the entire depth map is traversed.
3. The method for determining berth obstruction based on monocular depth ranging recognition according to claim 2, characterized in that: The specific process of judging whether the vehicle is in the target berth according to the intersection of the chassis quadrilateral of the vehicle and the quadrilateral of the target berth includes: Calculating the intersection area of the vehicle's chassis quadrilateral and the target berth quadrilateral; If the intersection area is greater than a preset threshold, it is determined that the vehicle is in the target parking space.
4. The method for determining berth obstruction based on monocular depth ranging recognition according to claim 3 is characterized in that: The process of separating the vehicle from the obstruction according to the depth ranging information and judging whether the vehicle is obstructed according to the pixel ratio of the separated obstruction and the vehicle includes: By utilizing the different depth information of the front and rear objects, when a vehicle and an obstruction exist in the image, the vehicle is separated from the obstruction according to the depth ranging information; The pixel area ratio of the obstruction and the vehicle is calculated, and when the pixel area ratio is greater than a preset threshold, it is determined that the vehicle is obstructed, otherwise it is determined that the vehicle is not obstructed.
5. The method for determining berth obstruction based on monocular depth ranging recognition according to claim 4, characterized in that: The process of extracting the edge of the occluder on the target berth and judging whether the target berth is blocked according to the pixel ratio of the occluder to the target berth includes: When there is an obstruction in the target berth area, extract edge information of the obstruction in the target berth according to the depth ranging information, and use the erosion morphology algorithm to denoise the edge information; Fitting the edge of the occluder into a closed figure through contour search and a distance-based closing algorithm; The ratio of the pixel area of the closed figure to the pixel area of the target berth is calculated, and when the ratio is greater than a preset threshold, it is determined that the target berth is blocked, otherwise it is determined that the target berth is not blocked.
6. The method for determining berth obstruction based on monocular depth ranging recognition according to any one of claims 1 to 5, characterized in that: Also includes: If the target berth is not blocked, clustering calculation is performed on the depth map using a clustering algorithm, and objects in the map are classified into three categories: far, medium, and near according to the depth information; Calculating a connected domain of a nearby object, and calculating average depth information of the nearby object based on the connected domain; If the average depth of the nearby object is greater than a specified threshold, and the ratio of the object area to the image area is greater than a preset threshold, it is determined that the image is occluded; otherwise, it is determined that the image is not occluded.
7. A berth occlusion judgment system based on monocular depth ranging recognition, characterized in that: The method for determining berth obstruction based on monocular depth ranging recognition as claimed in any one of claims 1 to 6 comprises: A depth image acquisition module, used to acquire an image of a target berth area and extract depth ranging information from the image; A vehicle model detection module, used to perform vehicle detection on the image based on a preset target detection model; A vehicle presence determination module, for determining whether the vehicle is in the target berth according to the intersection of the chassis quadrilateral of the vehicle and the quadrilateral of the target berth when a vehicle is detected; A vehicle shielding separation module, used for separating the vehicle from the shielding object according to the depth ranging information when the vehicle is located in the target berth; A vehicle occlusion judgment module, used to judge whether the vehicle is occluded according to the pixel ratio of the separated occluding object and the vehicle; The parking space blocking judgment module is used to extract the edge of the blocking object for the target parking space when no vehicle is detected, and judge whether the target parking space is blocked according to the pixel ratio of the blocking object to the target parking space.
8. The berth obstruction judgment system based on monocular depth ranging recognition according to claim 7 is characterized in that: It also includes an image occlusion judgment module, which is specifically used to: When the target berth is not blocked, clustering calculation is performed on the depth map using a clustering algorithm, and objects in the map are classified into three categories: far, medium, and near according to depth information; Calculating a connected domain of a nearby object, and calculating average depth information of the nearby object based on the connected domain; If the average depth of the nearby object is greater than a specified threshold, and the ratio of the object area to the image area is greater than a preset threshold, it is determined that the image is occluded; otherwise, it is determined that the image is not occluded.
9. The berth obstruction judgment system based on monocular depth ranging recognition according to claim 7, characterized in that: The vehicle shielding separation module is specifically used for: By utilizing the different depth information of the front and rear objects, when a vehicle and an obstruction exist in the image, the vehicle is separated from the obstruction according to the depth ranging information; The pixel area ratio of the obstruction and the vehicle is calculated, and when the pixel area ratio is greater than a preset threshold, it is determined that the vehicle is obstructed, otherwise it is determined that the vehicle is not obstructed.
10. The berth obstruction judgment system based on monocular depth ranging recognition according to claim 9, characterized in that: The vehicle occlusion judgment module is specifically used for: When there is an obstruction in the target berth area, extract edge information of the obstruction in the target berth according to the depth ranging information, and use the erosion morphology algorithm to denoise the edge information; Fitting the edge of the occluder into a closed figure through contour search and a distance-based closing algorithm; The ratio of the pixel area of the closed figure to the pixel area of the target berth is calculated, and when the ratio is greater than a preset threshold, it is determined that the target berth is blocked, otherwise it is determined that the target berth is not blocked.