Road target detection system, method and device

Abstract

The invention relates to the technical field of traffic management, in particular to a road target detection system, method and device, and the method comprises the steps: collecting a vehicle image of a roadside parking region; analyzing the vehicle image data to detect vehicle information in the parking space; based on the vehicle information, blind area vehicles generated due to shielding of front and rear vehicles are recognized and marked; recording parking information of a vehicle in a parking space in a blind area, wherein the parking information comprises mark information, a position and parking time; after the vehicle in the blind area is driven out of the parking space, whether the marked vehicle is still in a shielding state or not is checked; if the marked vehicle is still in the shielding state, continuously acquiring feature information of the marked vehicle; searching and matching corresponding vehicles in other cameras in a preset range based on the feature information to obtain a target vehicle; and obtaining license plate information based on the target vehicle through a license plate recognition technology, and replacing the mark information in the parking information with the license plate information.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic management, and particularly to a road target detection system, method and device. Background Art

[0002] The roadside parking space recognition technology is an intelligent system that uses a variety of sensors and image processing algorithms to automatically detect and locate available parking spaces on urban roadsides. This technology usually combines advanced technologies such as computer vision, machine learning, and the Internet of Things (IoT). By using cameras or other sensors (such as lidar, ultrasonic sensors) installed on vehicles to obtain environmental data, and then analyzing this data to determine whether there are vacant parking spaces.

[0003] However, due to the cameras being placed at a high position, if two vehicles are too close to each other, the license plate information cannot be obtained, thus the vehicle identity cannot be determined, resulting in statistical omissions and reducing the management efficiency of roadside parking spaces. Summary of the Invention

[0004] The purpose of the present invention is to provide a road target detection system, method and device, aiming to solve the problem of vehicle parking recognition in the case of license plate occlusion during the recognition of roadside parked vehicles, thereby improving the management efficiency of roadside parking spaces.

[0005] To achieve the above purpose, in a first aspect, the present invention provides a road target detection method, including collecting vehicle images of a roadside parking area;

[0006] Analyzing the vehicle image data to detect vehicle information within the parking space;

[0007] Identifying blind - area vehicles caused by front - and - rear vehicle occlusion based on the vehicle information, and marking them;

[0008] Recording the parking information of the blind - area vehicles in the parking space, where the parking information includes marking information, position, and parking time;

[0009] After the blind - area vehicle drives out of the parking space, checking whether the marked vehicle is still in an occluded state; if the marked vehicle is still in an occluded state, continuously obtaining the feature information of the marked vehicle;

[0010] Searching for and matching the corresponding vehicle in other cameras within a preset range based on the feature information to obtain the target vehicle;

[0011] Obtaining the license plate information based on the target vehicle through license plate recognition technology, and replacing the marking information in the parking information with the license plate information.

[0012] Among them, the specific steps of collecting vehicle images of the roadside parking area include:

[0013] Set the first time interval to automatically collect images to obtain the first image sequence;

[0014] When a vehicle is detected entering the parking space based on the first image sequence, capture the second image sequence using the second time interval;

[0015] Upload the second image sequence to the cloud for storage to obtain vehicle images.

[0016] Among them, the specific steps for analyzing the vehicle image data to detect the vehicle information in the parking space include:

[0017] Preprocess the collected vehicle images;

[0018] Exclude objects that have been stationary for a long time from the preprocessed vehicle images through the background subtraction algorithm, and only retain moving or temporarily parked vehicles;

[0019] Use the difference analysis between consecutive frames to detect whether a vehicle enters the parking space to determine whether each parking space is occupied, and update the parking space status information.

[0020] Among them, the specific steps for excluding objects that have been stationary for a long time from the preprocessed vehicle images through the background subtraction algorithm and only retaining moving or temporarily parked vehicles include:

[0021] When there is no obvious activity in the parking lot or monitoring area, collect images as background samples;

[0022] Establish a background model based on the background samples;

[0023] Compare each newly collected image frame by frame with the current background model to calculate the difference;

[0024] Set a threshold according to the degree of difference. Pixels exceeding this threshold are considered foreground parts. Objects that appear in the same position for consecutive frames and have not undergone significant displacement are regarded as static objects and excluded from the foreground;

[0025] Screen the detected foreground objects based on the known parking space size and vehicle shape features to exclude abnormal objects.

[0026] Among them, the specific steps for using the difference analysis between consecutive frames to detect whether a vehicle enters the parking space to determine whether each parking space is occupied include:

[0027] Assign a unique tracking ID to each detected vehicle to continuously track its position and status changes during the entire parking period;

[0028] Combine the historical movement trajectories of the vehicle to predict its future moving direction;

[0029] When it is detected that the vehicle has completely entered the parking space, immediately update the status information of the parking space and record the relevant timestamp in the system.

[0030] Among them, the specific steps of identifying blind - area vehicles caused by front - and - rear vehicle occlusion based on vehicle information and marking them include:

[0031] By analyzing the relative position relationship between adjacent vehicles, determine whether there is an occlusion situation;

[0032] For vehicles that have been occluded but have been successfully tracked before, their current position can be predicted by analyzing their historical movement trajectories;

[0033] Check the movement pattern of the vehicle between consecutive frames. If it is found that a certain tracking ID disappears in some frames but reappears in subsequent frames, it can be inferred that the vehicle was once in the blind area;

[0034] Add a mark to the vehicle determined to be in the blind area.

[0035] Among them, the specific steps of recording the parking information of the blind - area vehicle in the parking space include:

[0036] Obtain the marking information;

[0037] When the vehicle is marked for the first time, set the corresponding parking - space status to occupied and record the current time as the entry time;

[0038] Start timing from the entry time until the vehicle leaves the parking space, and calculate the total parking time.

[0039] Among them, the specific steps of, after the blind - area vehicle drives out of the parking space, checking whether the marked vehicle is still in the occluded state; if the marked vehicle is still in the occluded state, continuously obtaining the feature information of the marked vehicle include:

[0040] When the vehicle starts to move, check whether the vehicle is the previously marked blind - area vehicle;

[0041] Continue to monitor the movement of the vehicle in real - time. Once the vehicle is completely visible and it is confirmed that it is no longer in the occluded state, record the timestamp and license - plate information at this moment; if the vehicle is still occluded, capture the feature information of the vehicle, and the feature information includes color, model outline, and pattern.

[0042] In a second aspect, the present invention also provides a road target detection device, including:

[0043] An image acquisition module, a vehicle detection module, a blind - area vehicle identification module, a parking - information statistics module, a feature extraction module, a matching module, and an update module;

[0044] The image acquisition module is used to acquire vehicle images of the roadside parking area;

[0045] The vehicle detection module is used to analyze vehicle image data to detect vehicle information within the parking space;

[0046] The blind - area vehicle identification module is used to identify blind - area vehicles caused by the occlusion of front and rear vehicles based on vehicle information and mark them;

[0047] The parking information statistics module is used to record the parking information of blind - area vehicles in the parking space, and the parking information includes mark information, position, and parking time;

[0048] The feature extraction module is used to check whether the marked vehicle is still in an occluded state after the blind - area vehicle drives out of the parking space; if the marked vehicle is still in an occluded state, continuously obtain the feature information of the marked vehicle;

[0049] The matching module is used to search for and match the corresponding vehicle in other cameras within a preset range based on the feature information to obtain the target vehicle;

[0050] The update module is used to obtain license - plate information based on the target vehicle through license - plate recognition technology and replace the mark information in the parking information with the license - plate information.

[0051] In a third aspect, the present invention further provides a road target detection system, including the described road target detection device.

[0052] A road target detection system, method and device of the present invention install an intelligent camera or a sensor network in a designated roadside parking area to periodically or real-time collect high-definition images of parked vehicles. These images will serve as the basic data for subsequent analysis. Computer vision algorithms and deep learning models are used to process and analyze the collected vehicle image data. For vehicles in the visual blind spots formed by the occlusion of the front and rear vehicles, the system uses algorithms to identify them and assigns a unique label to each such blind spot vehicle. For all labeled blind spot vehicles, the system will record their parking information in detail, including but not limited to: label number, specific parking space position coordinates, timestamp of entering the parking space, etc. When a labeled vehicle drives out of the parking space, the system will automatically trigger an inspection to confirm whether the vehicle has completely left the occlusion area. If it is found that the vehicle is still partially occluded (for example, moving slowly but not completely leaving), the characteristic information of the vehicle will continue to be collected. To ensure that the vehicle can be effectively tracked even when it moves into the coverage range of other cameras, the system will establish a cooperation mechanism among multiple cameras within a preset range. According to the characteristic information of the vehicle, such as body color, vehicle type, etc., search and match the same vehicle in the images taken from different angles. Once the identity of the target vehicle is determined, the system will start the high-precision license plate recognition technology to accurately obtain the license plate number. Finally, the system replaces the temporary label information assigned to the vehicle with the actual license plate information to complete the closed-loop of the entire detection process.

[0053] Through the above method, the problem of vehicle parking recognition in the case of license plate occlusion during the recognition of roadside parked vehicles can be solved, thereby improving the management efficiency of roadside parking spaces. Description of the Drawings

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0055] Figure 1 It is a flowchart of a road target detection method of the present invention.

[0056] Figure 2 It is a flowchart of collecting vehicle images in the roadside parking area of the present invention.

[0057] Figure 3 It is a flowchart of analyzing vehicle image data to detect vehicle information in the parking space of the present invention.

[0058] Figure 4It is a flowchart of the present invention for excluding long-term stationary objects from the preprocessed vehicle images through the background subtraction algorithm and only retaining moving or temporarily parked vehicles.

[0059] Figure 5 It is a flowchart of the present invention for detecting whether a vehicle enters a parking space by analyzing the differences between consecutive frames to determine whether each parking space is occupied.

[0060] Figure 6 It is a flowchart of the present invention for identifying blind spot vehicles caused by front and rear vehicle occlusion based on vehicle information and marking them.

[0061] Figure 7 It is a flowchart of the present invention for recording the parking information of blind spot vehicles located in parking spaces.

[0062] Figure 8 It is a flowchart of the present invention for checking whether the marked vehicle is still in an occluded state after the blind spot vehicle drives out of the parking space; if the marked vehicle is still in an occluded state, continuously obtain the characteristic information of the marked vehicle. Detailed implementation manners

[0063] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0064] The first embodiment

[0065] Please refer to Figures 1 to 8 , the present invention provides a road target detection method, including:

[0066] S101 Collect vehicle images of the roadside parking area;

[0067] The specific steps include:

[0068] S201 Set the first time interval to automatically collect images to obtain the first image sequence;

[0069] Install a high-definition camera to ensure that it covers all target parking spaces, and adjust parameters such as focal length and aperture to obtain clear and stable images. Calibrate the camera to ensure that the position of each parking space in the image is fixed and there is no obvious distortion. In the case of no vehicle movement, collect a series of static images as the basis of the background model for subsequent inter-frame difference analysis.

[0070] According to the specific conditions of the parking lot (such as traffic flow, lighting conditions, etc.), set a reasonable first time interval (for example, capture one frame of image per second or every few seconds). Start the timer and automatically trigger the camera to capture images at the set time interval to form the first image sequence. The first image sequence is mainly used to monitor the state changes of the overall parking area and capture potential vehicle entry or exit events.

[0071] S202 When it is detected based on the first image sequence that a vehicle enters a parking space, capture the second image sequence using a second time interval;

[0072] Perform pixel-by-pixel comparison on consecutive frames in the first image sequence and calculate the difference between frames. Set a reasonable threshold, and pixels exceeding this threshold are considered the foreground part, indicating that there is an object moving. Apply morphological processing (such as dilation, erosion) to remove noise points and smooth the contours to ensure clear boundaries of the detected objects.

[0073] Based on the known dimensions of the parking space and the shape characteristics of the vehicle, screen the detected connected regions and exclude small or abnormal objects that do not meet the conditions (such as pedestrians, pets, etc.). Confirm whether the connected region is within or near the parking space to preliminarily determine whether a vehicle enters or leaves the parking space.

[0074] Once it is detected that a vehicle enters a parking space, immediately switch to a shorter second time interval (such as multiple frames per second) and start the high-frequency image capture mode. Capture the second image sequence to record in detail the process of the vehicle from entering the parking space to coming to a complete stop.

[0075] The second image sequence is used to refine the tracking of the vehicle's dynamic changes to ensure capturing high-quality vehicle images, especially license plate information.

[0076] S203 Upload the second image sequence to the cloud for storage to obtain the vehicle images.

[0077] Perform preprocessing on each frame in the second image sequence, including operations such as denoising, contrast enhancement, and color correction, to improve the image quality and the accuracy of subsequent analysis.

[0078] Since a multi-camera system is used, it is necessary to synchronize the image timestamps of each camera to ensure data consistency. Upload the processed images to the cloud storage platform through a secure network channel to ensure the security and integrity of the transmission. Use encryption technology to protect privacy information and prevent unauthorized access.

[0079] S102 Analyze the vehicle image data to detect vehicle information within the parking space;

[0080] The specific steps include:

[0081] S301 Perform preprocessing on the collected vehicle images;

[0082] Apply a denoising algorithm (such as Gaussian filtering) to remove random noise in the image and improve image clarity. Then use contrast stretching and histogram equalization techniques to enhance the visual effect of the image and make vehicle features more obvious. After that, convert the image from the RGB color space to a color space more suitable for vehicle detection (such as HSV or YCbCr) to improve detection accuracy.

[0083] In S302, for the preprocessed vehicle image, exclude objects that have been stationary for a long time through a background subtraction algorithm, and only retain moving or temporarily parked vehicles;

[0084] The specific steps include:

[0085] In S401, when there is no obvious activity in the parking lot or monitoring area, collect an image as a background sample;

[0086] Select a time period with less vehicle flow in the parking lot or monitoring area (such as late at night or early in the morning) to ensure that the background sample is as pure as possible. Continuously collect multiple frames of images to obtain a stable background sample and reduce errors caused by short-term interferences (such as pedestrians, animals).

[0087] In S402, establish a background model based on the background sample;

[0088] Use methods such as the average method, median method, or Gaussian mixture model (GMM) to construct the background model.

[0089] For scenarios with large environmental changes, consider using an adaptive background modeling algorithm, such as ViBe (Visual Background Extractor), to improve the robustness of the model.

[0090] In S403, compare each newly collected image with the current background model pixel by pixel and calculate the difference;

[0091] For each newly collected image, compare it with the current background model pixel by pixel and calculate the difference between the two. The difference can be calculated through a simple subtraction operation or a more complex distance metric (such as Euclidean distance, Mahalanobis distance).

[0092] In S404, set a threshold based on the degree of difference. Pixels exceeding this threshold are considered the foreground part. For objects that appear in the same position in multiple consecutive frames and do not undergo significant displacement, they are regarded as static objects and excluded from the foreground;

[0093] Set a reasonable difference threshold according to the actual scenario and requirements. Pixels exceeding this threshold are considered the foreground part. For objects that appear at the same position in multiple consecutive frames and do not undergo significant displacement, they are regarded as static objects (such as trees, street lamps) and excluded from the foreground. Apply morphological operations such as dilation and erosion to remove noise points and smooth the contours to ensure clear boundaries of the detected objects.

[0094] S405 Screen the detected foreground objects based on the known parking space size and vehicle shape characteristics to exclude abnormal objects.

[0095] According to the known parking space size and vehicle shape characteristics, screen the detected foreground objects to exclude small or abnormal objects that do not meet the conditions (such as pedestrians, pets, etc.). Use the vehicle contour matching algorithm to further verify whether the detected objects conform to the vehicle shape characteristics to ensure that only real vehicles are retained.

[0096] S303 Use the difference analysis between consecutive frames to detect whether a vehicle enters the parking space to determine whether each parking space is occupied and update the parking space status information.

[0097] The specific steps include:

[0098] S501 Assign a unique tracking ID to each detected vehicle to continuously track its position and status changes during the entire parking period;

[0099] When a new vehicle is first detected, assign a unique tracking ID to it to ensure that its position and status changes can be continuously tracked during the entire parking period. The tracking ID can help the system distinguish different vehicles even if they are similar in appearance.

[0100] S502 Combine the historical movement trajectory of the vehicle to predict its future movement direction;

[0101] Combine the historical movement trajectory of the vehicle to predict its future movement direction to assist subsequent detection and tracking work. This helps to identify potential parking space occupancy in advance and optimize the parking management strategy.

[0102] S503 When it is detected that the vehicle has completely entered the parking space, immediately update the status information of this parking space and record the relevant timestamp in the system.

[0103] When it is detected that the vehicle has completely entered the parking space, immediately update the status of this parking space from "idle" to "occupied" and record the relevant entry timestamp in the system. Similarly, when the vehicle leaves the parking space, update the status to "idle" and record the departure timestamp.

[0104] Record in detail each event of a vehicle entering and leaving a parking space, including the license plate number, entry time, and departure time, to form a complete activity log. The log is not only used for internal management but can also be provided to vehicle owners to query parking records, enhancing the service experience.

[0105] S103 identifies blind spot vehicles caused by the occlusion of the front and rear vehicles based on vehicle information and marks them.

[0106] The specific steps include:

[0107] S601 determines whether there is an occlusion situation by analyzing the relative position relationship between adjacent vehicles.

[0108] Capture images of the parking lot area using multiple cameras at different angles to ensure that even if there is vehicle occlusion, partial information of blind spot vehicles can be obtained from other angles. Combine the data of the video surveillance system and other sensors (such as radar, ultrasonic sensors) to improve the accuracy of detecting occlusion situations.

[0109] Use the parking space segmentation algorithm and vehicle contour matching technology to accurately determine the position of vehicles in each parking space. Build a geometric model to predict which vehicles will be occluded by other vehicles based on the size and position of the vehicles. Analyze the relative position relationship between adjacent vehicles, especially whether the projected areas of them in the image overlap. If significant spatial overlap is found between two or more vehicles, it is preliminarily judged that there is an occlusion situation.

[0110] S602 For vehicles that have been occluded but have been successfully tracked before, their current position can be predicted by analyzing their historical movement trajectories.

[0111] Consult the historical movement trajectory of the vehicle in the database to understand its past movement patterns and staying habits.

[0112] Based on the timestamps and paths of the vehicle entering and leaving the parking space in the recent few times, speculate on the current position of the vehicle. Apply machine learning or deep learning algorithms to establish a dynamic prediction model for vehicle movement. The model can adjust prediction parameters according to factors such as time, date, and weather to improve prediction accuracy.

[0113] Check the movement patterns of vehicles between consecutive frames and conduct multi-frame correlation analysis in combination with historical trajectory data. If it is found that a certain tracking ID disappears in some frames but reappears in subsequent frames, it can be speculated that the vehicle was once in the blind spot, and its specific position can be predicted using historical trajectory data.

[0114] S603 Check the movement patterns of vehicles between consecutive frames. If it is found that a certain tracking ID disappears in some frames but reappears in subsequent frames, it can be speculated that the vehicle was once in the blind spot.

[0115] For each vehicle being tracked, continuously record its movement patterns in consecutive frames, including characteristics such as speed, direction, and acceleration. When a certain tracking ID disappears in some frames, immediately trigger the anomaly detection mechanism to initiate further analysis. Considering the property that a vehicle appears at different positions at different time points, establish spatio-temporal association rules to assist the matching process. If it is found that a certain tracking ID disappears in some frames and reappears in subsequent frames, it is speculated that the vehicle was once in a blind area, and try to find more information from camera or sensor data from other perspectives.

[0116] S604 Add a mark to the vehicle determined to be in the blind area.

[0117] Add special marks or labels, such as color coding, symbol identification, etc., to the vehicles determined to be in the blind area to facilitate quick identification by management personnel. Mark these vehicles in the system interface to highlight their special status and remind management personnel to pay attention. Store all relevant information about these blind area vehicles (such as estimated position, license plate number, entry time, etc.) as the basis for further analysis.

[0118] Through the above steps, the process of S103 identifying blind area vehicles based on vehicle information and marking them not only ensures that all vehicles can be attended to and managed, but also improves the intelligence level and service quality of the entire parking management system.

[0119] S104 Record the parking information of the blind area vehicle in the parking space, where the parking information includes marking information, position, and parking time;

[0120] The specific steps include:

[0121] S701 Obtain the marking information;

[0122] Confirm the source of the mark of the blind area vehicle to ensure that it is automatically identified and added by the system or marked through manual intervention. For automatically marked vehicles, check the timestamp and reason for marking (such as occlusion detection, historical trajectory prediction, etc.) to verify the validity and accuracy of the mark.

[0123] S702 When the vehicle is marked for the first time, set the corresponding parking space status to occupied and record the current time as the entry time;

[0124] When the vehicle is first marked as a blind spot vehicle, immediately update the status of the parking space from "available" to "occupied", and record this change in the system. Updating the parking space status not only helps to monitor the usage of the parking lot in real time, but also provides accurate data support for other functional modules (such as parking fee calculation, route planning). Precisely record the current time as the "entry time" of the vehicle, which serves as the starting point for parking billing and other statistical analyses. The entry timestamp should be accurate to the second to ensure the accuracy of subsequent calculations and reduce disputes caused by time errors.

[0125] S703 Start timing from the entry time until the vehicle leaves the parking space, and calculate the total stay time.

[0126] Starting from the entry time, start the timer and continuously monitor the position change of the vehicle until it completely leaves the parking space. When it is detected that the vehicle has completely left the parking space, immediately record the current time as the "departure time", which serves as the end point for parking billing and other statistical analyses. Calculate the total stay time of the vehicle in the parking space based on the entry time and the departure time. The total stay time can be used for functions such as parking fee calculation and overtime reminder to ensure reasonable and transparent charging.

[0127] For an object that stays at a certain position for a long time but is not a fixed facility, start a special review process to determine whether it is illegally parked or in other special situations. If any abnormal behavior (such as illegal occupation, overtime parking, etc.) is found, immediately trigger the alarm mechanism and notify the relevant personnel for handling.

[0128] S105 After the blind spot vehicle drives out of the parking space, check whether the marked vehicle is still in a blocked state; if the marked vehicle is still in a blocked state, continuously obtain the characteristic information of the marked vehicle;

[0129] The specific steps include:

[0130] S801 When the vehicle starts to move, check whether the vehicle is the previously marked blind spot vehicle;

[0131] When it is detected by a camera or sensor (such as a geomagnetic inductor, ultrasonic sensor) that a vehicle starts to move, trigger the vehicle departure detection process. Check whether the vehicle has been marked as a blind spot vehicle by the system to confirm its special status. Update the status of the marked vehicle from "parked" to "moving", and start the subsequent detection process. Record the timestamp when the vehicle starts to move, which serves as the basis for subsequent analysis.

[0132] S802 Continue to monitor the movement of the vehicle in real time. Once the vehicle is completely visible and confirmed to be no longer in a blocked state, record the timestamp and license plate information at this moment; if the vehicle is still blocked, capture the characteristic information of the vehicle, and the characteristic information includes color, model outline, and pattern.

[0133] Start the high-frequency image acquisition mode and continue to monitor the vehicle's movement in real time to ensure that no critical moments are missed. Predict the vehicle's future movement direction using its historical movement trajectory to assist subsequent detection and tracking work. Regularly check whether the vehicle has moved out of the occlusion area, and ensure the accuracy of the judgment through multi-frame correlation analysis and multi-view verification. Once the vehicle is fully visible, immediately confirm that it is no longer in an occluded state and record the timestamp of this moment.

[0134] When the vehicle is fully visible, quickly perform license plate recognition operations, read the license plate number and record relevant information. For blurred or low-quality license plate images, apply advanced image processing techniques to enhance clarity to improve the recognition success rate.

[0135] Even if the vehicle is partially occluded, try to capture basic features such as the vehicle's color, model outline, etc. And specifically extract and identify prominent markers on the vehicle body, such as stickers, decorative strips or other unique patterns, which help subsequent matching and identification.

[0136] Try to capture detailed features such as the vehicle's surface pattern, logo, wheel style, etc. to form a more comprehensive vehicle profile. If necessary, use an infrared camera or night vision function to obtain more details under night or low-light conditions.

[0137] Try to obtain more information from cameras at different angles to make up for the deficiencies of a single perspective. Integrate data from multiple cameras to build a more complete vehicle feature model.

[0138] Store the captured feature information in the central database, associate it with previous records to form a complete vehicle profile. Update the vehicle's latest position and status information to ensure that all data is consistent. Record in detail each event of vehicle status change, including the timestamp, feature information and position change, to form a complete activity log. The log is not only used for internal management but also can be provided to the vehicle owner to query parking records to enhance the service experience.

[0139] S106 Search for and match the corresponding vehicle in other cameras within a preset range based on the feature information to obtain the target vehicle;

[0140] According to the layout of the parking lot or roadside parking area, determine which cameras are within the preset range and establish the topological relationship between the cameras. Ensure that these cameras cover the vehicle's movement path, especially at key positions such as the entrance and exit of the parking lot and turning points. Ensure that all image data and vehicle feature information collected by the cameras are stored in the central database for subsequent query.

[0141] Normalize the images from different cameras to eliminate the differences caused by factors such as lighting conditions and shooting angles. Apply color space conversion (such as from RGB to HSV) to improve the accuracy of feature matching.

[0142] Extract stable feature information from the original images of the marked vehicles, including but not limited to color, vehicle model contour, license plate number, special markers, etc. Integrate the feature information from multiple cameras to construct a more comprehensive vehicle feature model. Utilize the property that the vehicle appears at different positions at different time points to establish spatio-temporal association rules to assist the matching process. Apply machine learning or deep learning algorithms to establish a dynamic prediction model for vehicle movement. According to the basic features of the vehicle (such as color, size), quickly filter the camera data within a preset range to reduce the number of potential matching objects. Combine the vehicle's moving speed and direction to predict the area it will reach, further reducing unnecessary computational workload. Set a reasonable search area based on the vehicle's last appearance position and moving direction to avoid blind searching within the entire parking lot.

[0143] Use a trained deep learning model (such as Convolutional Neural Network CNN) to perform more accurate matching on the candidate vehicles after preliminary screening. The deep learning model can identify complex visual features and improve the accuracy of matching. Calculate the similarity score between each candidate vehicle and the target vehicle, and the one with the highest score is taken as the most likely matching result. Set a reasonable similarity threshold, and only when the score of the candidate vehicle exceeds this threshold is it considered that a matching vehicle has been found.

[0144] For the candidate vehicles with high scores, check their performance in multiple consecutive frames to ensure it is not a coincidental similarity.

[0145] Multi-frame confirmation can exclude false matches caused by temporary occlusion or other interference factors. Instead of relying on a single feature, multiple features (such as license plate number, vehicle body details) are comprehensively reviewed to improve the matching accuracy.

[0146] Once a matching vehicle is found, update its movement path in the entire parking lot or road network. Record the latest position and status information of the vehicle to ensure that all relevant data is consistent. Send a text message or push notification containing the parking location, parking time, and fee to the vehicle owner for confirmation.

[0147] S107 obtains the license plate information based on the target vehicle through license plate recognition technology, and replaces the marked information in the parking information with the license plate information.

[0148] Ensure that the target vehicle has been successfully matched among other cameras within the preset range. Check the similarity score of the matching result to ensure that it exceeds the set threshold to guarantee the accuracy of the match. Select the clearest and best-angled image from multiple cameras for license plate recognition. Apply a denoising algorithm (such as Gaussian filtering) to remove random noise in the image and improve image clarity. Use contrast stretching and histogram equalization techniques to enhance the visual effect of the license plate area. Utilize license plate detection algorithms (such as Haar feature cascade classifier, deep learning models, etc.) to accurately locate the position of the license plate in the image. Separate the license plate from the background to form an independent image block for subsequent character recognition. Apply OCR technology to read the text on the license plate and extract the license plate number. For blurred or low-quality license plate images, apply advanced image processing techniques (such as super-resolution reconstruction, edge enhancement) to improve the recognition success rate. Perform format and logical verification on the recognized license plate number to ensure its validity and accuracy. If there is uncertainty in the recognition result, initiate the manual intervention option to allow the management personnel to make the final judgment.

[0149] For the situation where the license plate fails to be recognized successfully for a long time, initiate a special review process due to license plate damage or other special circumstances.

[0150] Second Embodiment

[0151] The present invention provides a road target detection device, including: an image acquisition module, a vehicle detection module, a blind area vehicle recognition module, a parking information statistics module, a feature extraction module, a matching module, and an update module; the image acquisition module is used to acquire vehicle images in the roadside parking area; the vehicle detection module is used to analyze vehicle image data to detect vehicle information in the parking space; the blind area vehicle recognition module is used to identify blind area vehicles caused by front and rear vehicle occlusion based on the vehicle information and mark them; the parking information statistics module is used to record the parking information of the blind area vehicle in the parking space, and the parking information includes marking information, position, and parking time; the feature extraction module is used to check whether the marked vehicle is still in the occluded state after the blind area vehicle drives out of the parking space; if the marked vehicle is still in the occluded state, continuously obtain the feature information of the marked vehicle; the matching module is used to search for and match the corresponding vehicle in other cameras within the preset range based on the feature information to obtain the target vehicle; the update module is used to obtain the license plate information based on the target vehicle through license plate recognition technology and replace the marking information in the parking information with the license plate information.

[0152] In this embodiment, the image acquisition module captures and provides high-definition real-time vehicle image data, covering the entire roadside parking area. These images not only provide the basic data for subsequent analysis but also ensure that there are no blind spots in the monitoring range of the system. The vehicle detection module deeply analyzes the static or dynamic image data provided by the image acquisition module, accurately detects the vehicles parked in the parking spaces, and collects key information about the vehicles, such as size, color, etc., for further processing. For those "blind spot vehicles" that are difficult to directly observe due to the occlusion of the front and rear vehicles, the blind spot vehicle recognition module plays an important role. It can infer the existence of vehicles in the blind spot based on the positional relationship of adjacent vehicles and the known vehicle information, and specially mark them to ensure that all vehicles can be accurately tracked. The parking information statistics module closely tracks the status changes of each parking space, and details the parking details of each vehicle marked as a blind spot vehicle. This includes but is not limited to the specific markings of the vehicle, the location coordinates, and the specific time point when it enters the parking space, so as to provide detailed data support for the management department. When a vehicle that was once marked as a blind spot is about to leave its parking space, the feature extraction module comes into play. It checks whether the vehicle is still under the occlusion of other vehicles; if it is confirmed that there is still occlusion, it continuously captures the unique appearance features of the vehicle, such as the body contour, logo pattern, etc., for subsequent identification work.

[0153] The task of the matching module is to use the information obtained from the feature extraction module to search within a preset geographical range for the presence of a target vehicle that matches it in the images captured by other cameras. Once the corresponding vehicle is found, its whereabouts can be determined to achieve continuous monitoring across regions. The update module introduces license plate recognition technology to accurately locate the target vehicle and read its license plate number. In this way, the system can replace the initially used temporary identification marking information with the real license plate information, complete the final update of the parking record, making each parking event traceable for subsequent query and management.

[0154] In summary, this road target detection device can not only effectively solve the blind spot problem in the traditional parking management system but also greatly improve the accuracy and timeliness of parking information, contributing an important force to the construction of urban intelligent transportation.

[0155] Third Embodiment

[0156] The present invention also provides a road target detection system, including the described road target detection device.

[0157] Based on the described road target detection device, this system further integrates a variety of intelligent components and service platforms to form a complete solution. Specifically, in addition to the visual information provided by the image acquisition module, the system can also integrate data from various sensing devices such as radar, lidar (LiDAR), and ultrasonic sensors. This enables the system to perceive the surrounding environment more comprehensively and maintain high-precision detection capabilities even under adverse weather conditions.

[0158] By leveraging a powerful cloud computing platform and edge computing technology, the system can quickly process massive data streams to achieve real-time target recognition, tracking, and warning. This means it can respond immediately to emergencies such as traffic accidents or illegal parking and notify relevant departments to take action in a timely manner.

[0159] The above-disclosed is only a preferred embodiment of the present invention. Of course, it cannot be used to limit the scope of the rights of the present invention. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. A road target detection method, characterized in that: include: Collect vehicle images in roadside parking areas; Analyzing vehicle image data to detect vehicle information within a parking space; Identify vehicles in blind spots caused by front and rear vehicles based on vehicle information and mark them; Recording parking information of vehicles in the blind spot at parking spaces, the parking information including marking information, location and parking time; After the blind spot vehicle exits the parking space, check whether the marked vehicle is still in the obstruction state; If the marked vehicle is still in an obstructed state, the characteristic information of the marked vehicle is continuously obtained; Based on the feature information, search and match the corresponding vehicle in other cameras within a preset range to obtain the target vehicle; The license plate information is obtained based on the target vehicle through the license plate recognition technology, and the license plate information replaces the marking information in the parking information.

2. A road target detection method as claimed in claim 1, characterized in that: The specific steps of collecting vehicle images in the roadside parking area include: Setting a first time interval to automatically capture images to obtain a first image sequence; capturing a second image sequence at a second time interval when a vehicle is detected to enter the parking space based on the first image sequence; The second image sequence is uploaded to the cloud storage to obtain the vehicle image.

3. A road target detection method as claimed in claim 2, characterized in that: The specific steps of analyzing the vehicle image data to detect the vehicle information in the parking space include: Preprocessing the collected vehicle images; The pre-processed vehicle images are subjected to background subtraction algorithms to exclude long-term stationary objects, and only moving or temporarily parked vehicles are retained; Using the difference analysis between consecutive frames, it is detected whether a vehicle enters the parking space to determine whether each parking space is occupied, and the parking space status information is updated.

4. A road target detection method as claimed in claim 3, characterized in that: The specific steps of eliminating long-term fixed objects by using a background subtraction algorithm on the pre-processed vehicle image and retaining only moving or temporarily parked vehicles include: When there is no obvious activity in the parking lot or monitoring area, collect images as background samples; Establish a background model based on the background samples; Compare each newly acquired image frame with the current background model pixel by pixel and calculate the difference; A threshold is set according to the degree of difference. Pixels exceeding this threshold are considered as foreground parts. Objects that appear in the same position for multiple consecutive frames and do not move significantly are considered as static objects and are excluded from the foreground. The detected foreground objects are screened based on the known parking space size and vehicle shape features to exclude abnormal objects.

5. A road target detection method as claimed in claim 4, characterized in that: The specific steps of using difference analysis between consecutive frames to detect whether a vehicle enters a parking space to determine whether each parking space is occupied include: Assign a unique tracking ID to each detected vehicle so that its location and status changes can be continuously tracked throughout the parking period; Combine the historical movement trajectory of the vehicle to predict its future movement direction; When it is detected that the vehicle has completely entered the parking space, the status information of the parking space is immediately updated and the relevant timestamp is recorded in the system.

6. A road target detection method as claimed in claim 5, characterized in that: The specific steps of identifying vehicles in the blind spot caused by the obstruction of the front and rear vehicles based on the vehicle information and marking them include: By analyzing the relative position relationship between adjacent vehicles, determine whether there is an occlusion; For vehicles that have been blocked but successfully tracked before, their current positions can be predicted by analyzing their historical motion trajectories; Check the movement pattern of the vehicle between consecutive frames. If a tracking ID disappears in some frames but reappears in subsequent frames, it can be inferred that the vehicle was once in the blind spot. Add markings to vehicles that are identified as being in the blind spot.

7. A road target detection method as claimed in claim 6, characterized in that: The specific steps of recording the parking information of the blind spot vehicle in the parking space include: Get tag information; When a vehicle is marked for the first time, the corresponding parking space status is set to occupied, and the current time is recorded as the entry time; The total stay time is calculated from the time of entry until the vehicle leaves the parking space.

8. A road target detection method as claimed in claim 7, characterized in that: After the vehicle in the blind spot drives out of the parking space, checking whether the marked vehicle is still in a blocked state; If the marked vehicle is still in an obstructed state, the specific steps of continuously acquiring the characteristic information of the marked vehicle include: When a vehicle starts to move, check whether the vehicle is a previously marked blind spot vehicle; Continue to monitor the movement of the vehicle in real time. Once the vehicle is fully visible and confirmed to be no longer obscured, record the timestamp and license plate information of this moment. If the vehicle is still obscured, capture the characteristic information of the vehicle, including color, model outline and pattern.

9. A road object detection device, applied to a road object detection method according to any one of claims 1 to 8, characterized in that: include: Image acquisition module, vehicle detection module, blind spot vehicle recognition module, parking information statistics module, feature extraction module, matching module and update module; The image acquisition module is used to acquire vehicle images in the roadside parking area; The vehicle detection module is used to analyze the vehicle image data to detect the vehicle information in the parking space; The blind spot vehicle identification module is used to identify vehicles in the blind spot caused by the obstruction of the front and rear vehicles based on the vehicle information and mark them; The parking information statistics module is used to record the parking information of vehicles in the blind spot parking spaces, wherein the parking information includes marking information, location and parking time; The feature extraction module is used to check whether the marked vehicle is still in an obstructed state after the blind spot vehicle drives out of the parking space; If the marked vehicle is still in an obstructed state, the characteristic information of the marked vehicle is continuously obtained; The matching module is used to search and match the corresponding vehicle in other cameras within a preset range based on the feature information to obtain the target vehicle; The updating module is used to obtain the license plate information based on the target vehicle through the license plate recognition technology, and replace the marking information in the parking information with the license plate information.

10. A road target detection system, characterized in that: It includes a road target detection device as described in claim 9.

Images