Parking lot high-order video identification system based on depth identification

Abstract

The invention discloses a parking lot high-order video recognition system based on depth recognition. The system comprises a video acquisition module, an image preprocessing module, a behavior analysis module, a parking space detection module, a license plate recognition module, a user interaction module and a data storage management module. According to the invention, a deep learning model based on the residual block is designed, the extraction capability of license plate features is optimized, the introduction of the residual block structure not only enhances the capture capability of the network for deep features, but also effectively alleviates the gradient disappearance problem through jump connection, so that the network can still maintain high performance while the depth is increased, and the method is suitable for the large-scale popularization and application. On this basis, an attention mechanism is further integrated, so that the neural network can autonomously adjust importance weights of all parts in input data, enhance information weights which are crucial to license plate recognition, and weaken other unnecessary parts, thereby guiding the network to be more focused on a license plate area, and improving the recognition accuracy of the license plate. And the attention of the feature extraction network to the license plate is obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of video recognition, and in particular to a parking lot high-position video recognition system based on depth recognition. Background Art

[0002] The parking lot high-altitude video recognition system is an advanced parking lot management technology. It uses high-altitude video cameras to accurately identify and manage vehicles and parking spaces in the parking lot. The system can detect the occupancy status of each parking space in the parking lot in real time and display this information on the user terminal to help car owners quickly find available parking spaces; the system can automatically identify the vehicle's license plate number and associate it with the owner's information to facilitate subsequent parking fee calculation and vehicle management; the system can provide car owners with intelligent navigation services based on their needs and the actual situation in the parking lot, helping them quickly find parking spaces and plan the best parking routes; the system can also detect abnormal behaviors in the parking lot, such as illegal parking, damaged vehicles, etc., and promptly notify the owner or parking lot manager for processing.

[0003] The existing parking lot high-position video recognition system has the following defects: First, the traditional license plate recognition algorithm mainly relies on the separation and extraction of color space to determine the license plate area. However, this method is often difficult to accurately extract the license plate position when the license plate color is not obvious or the lighting conditions vary greatly. For example, at night or in a dimly lit environment, the color characteristics of the license plate may become blurred, resulting in the algorithm being unable to effectively recognize it; and under strong light or backlight conditions, the color and brightness information of the license plate may also change significantly, further increasing the difficulty of recognition. Second, the license plate positioning method based on edge detection determines the edge of the license plate by extracting and connecting the edges of the license plate. When dealing with situations where there are obstructions on the license plate or the edges are not obvious, it is easy to make misjudgments. For example, when the license plate is When the license plate is partially occluded, the edge detection algorithm may not be able to accurately extract the complete edge of the license plate; when the edge of the license plate becomes blurred due to wear, stains, etc., the algorithm may mistakenly identify other objects as the edge of the license plate, resulting in positioning errors; third, the existing parking lot high-position video recognition system usually only uses a single trajectory feature to measure the similarity between the trajectory center and the trajectory. However, in complex parking lot monitoring scenarios, this method often cannot accurately identify abnormal trajectory patterns. In the case of dense traffic or intersecting trajectories, a single trajectory feature may not be able to effectively distinguish between normal and abnormal behaviors. When the pedestrian's speed, direction and other characteristics change significantly, the algorithm may also fail to accurately capture these changes, resulting in false alarms or missed alarms. Summary of the invention

[0004] The purpose of the present invention is to provide a parking lot high-position video recognition system based on depth recognition to solve the problems raised in the above background technology.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a parking lot high-position video recognition system based on depth recognition, comprising a video acquisition module, the video acquisition module is data-connected with an image preprocessing module, the image preprocessing module is data-connected with a behavior analysis module, a parking space detection module and a license plate recognition module, the behavior analysis module is data-connected with a user interaction module, and the user interaction module is data-connected with a data storage management module.

[0006] As a further technical solution of the present invention, the behavior analysis module includes a target detection module, a target tracking module, a trajectory feature extraction module, a skeleton feature extraction module, a feature fusion module, a pedestrian anomaly analysis module and a vehicle anomaly analysis module.

[0007] As a further technical solution of the present invention, the parking space detection module includes an occupancy detection module, a parking space status update module and a parking space allocation module, and the parking space detection module establishes a data connection with the user interaction module.

[0008] As a further technical solution of the present invention, the license plate recognition module includes a license plate positioning module, a license plate segmentation module, a character recognition module and a color recognition module, and the license plate recognition module establishes a data connection with the user interaction module.

[0009] As a further technical solution of the present invention, the license plate positioning module includes a residual block module, an attention mechanism module, an anchor frame optimization module and a positioning algorithm module.

[0010] As a further technical solution of the present invention, the attention mechanism module includes a channel attention module and a spatial attention module.

[0011] As a further technical solution of the present invention, the video acquisition module includes a high-position camera module, a video encoding module and a network communication module, and the network communication module establishes a data connection with the image preprocessing module.

[0012] As a further technical solution of the present invention, the image preprocessing module includes an image denoising module, an image enhancement module and an image scaling module.

[0013] As a further technical solution of the present invention, the user interaction module includes a user interface module, a rights management module and a user feedback module.

[0014] As a further technical solution of the present invention, the data storage management module includes a database module, a data backup module, a data query module and a data transmission module, and the data transmission module establishes a data connection with the user interaction module.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention designs a deep learning model based on residual blocks, optimizes the ability to extract license plate features, and the introduction of the residual block structure not only enhances the network's ability to capture deep-level features, but also effectively alleviates the gradient vanishing problem through jump connections, so that the network can still maintain high performance while increasing its depth. On this basis, an attention mechanism is further incorporated, so that the neural network can autonomously adjust the importance weights of each part of the input data, enhance the weights of information that is critical to license plate recognition, and weaken other unnecessary parts, thereby guiding the network to focus more on the license plate area, significantly improving the feature The feature extraction network pays attention to the license plate, and combines the color recognition and positioning technology with the targeted optimization of the anchor frame to locate the license plate area more accurately. It can maintain high accuracy and robustness even in complex and changeable scenes, which not only improves the efficiency of license plate recognition, but also greatly reduces the risk of misidentification. At the same time, it is designed to recognize abnormal pedestrian behavior based on multi-feature fusion. Combined with trajectory features and skeleton features, through multi-feature fusion, it realizes a more comprehensive and accurate recognition of abnormal pedestrian behavior. This fusion strategy not only improves the accuracy of recognition, but also enhances the system's adaptability to complex scenes and diverse behaviors, making abnormal behavior detection more reliable and stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a system structure diagram of the present invention;

[0017] Figure 2 It is a module architecture diagram of the image preprocessing module of the present invention;

[0018] Figure 3 It is a module architecture diagram of the behavior analysis module of the present invention;

[0019] Figure 4 The module architecture diagram of the license plate recognition module of the present invention;

[0020] Figure 5 The module architecture diagram of the license plate positioning module of the present invention;

[0021] Figure 6 Module architecture diagram of the attention mechanism module of the present invention;

[0022] Figure 7 A module architecture diagram of a data storage management module of the present invention;

[0023] Figure 8 It is a system flow chart of the present invention.

[0024] In the figure: 1. Video acquisition module; 11. High-position camera module; 12. Video encoding module; 13. Network communication module; 2. Image preprocessing module; 21. Image denoising module; 22. Image enhancement module; 23. Image scaling module; 3. Behavior analysis module; 31. Target detection module; 32. Target tracking module; 33. Trajectory feature extraction module; 34. Skeleton feature extraction module; 35. Feature fusion module; 36. Pedestrian anomaly analysis module; 37. Vehicle anomaly analysis module; 4. Parking space detection module; 41. Occupancy detection module; 42. Parking space status update module; 43. Parking space allocation module Block; 5. License plate recognition module; 51. License plate positioning module; 511. Residual block module; 512. Attention mechanism module; 5121. Channel attention module; 5122. Spatial attention module; 513. Anchor frame optimization module; 514. Positioning algorithm module; 52. License plate segmentation module; 53. Character recognition module; 54. Color recognition module; 6. User interaction module; 61. User interface module; 62. Authority management module; 63. User feedback module; 7. Data storage management module; 71. Database module; 72. Data backup module; 73. Data query module; 74. Data transmission module. DETAILED DESCRIPTION

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

[0026] Please refer to the attached Figure 1 -Attached Figure 8, an embodiment of the present invention: a parking lot high-position video recognition system based on depth recognition, including a video acquisition module 1, the video acquisition module 1 is data-connected with an image preprocessing module 2, the image preprocessing module 2 is data-connected with a behavior analysis module 3, a parking space detection module 4 and a license plate recognition module 5, the behavior analysis module 3 is data-connected with a user interaction module 6, and the user interaction module 6 is data-connected with a data storage management module 7; the behavior analysis module 3 includes a target detection module 31, a target tracking module 32, a trajectory feature extraction module 33, a skeleton feature extraction module 34, a feature fusion module 35, a pedestrian anomaly analysis module 36 and a vehicle anomaly analysis module 37, the target detection module 31 is used to detect pedestrians through YO The LOv4 algorithm performs pedestrian target detection. The target tracking module 32 is used to track the movement trajectory of the detection target. The trajectory feature extraction module 33 is used to extract the movement trajectory of the pedestrian target. The skeleton feature extraction module 34 is used for the skeleton feature of the target. The feature fusion module 35 is used to fuse the trajectory feature and the skeleton feature. The pedestrian abnormality analysis module 36 is used to analyze the abnormal behavior of pedestrians. The vehicle abnormality analysis module 37 is used to identify the abnormal behavior of vehicles, such as going against the flow and speeding. The parking space detection module 4 includes an occupancy detection module 41, a parking space status update module 42 and a parking space allocation module 43, and the parking space detection module 4 establishes a data connection with the user interaction module 6. The occupancy detection module 41 is used to detect whether the parking space is occupied, and the parking space status update module 42 is used to detect whether the parking space is occupied. Module 42 is used to update the status information of parking spaces in real time, such as vacant, occupied, etc., and parking space allocation module 43 is used to allocate appropriate parking spaces to vehicles according to the parking space conditions in the parking lot; license plate recognition module 5 includes license plate positioning module 51, license plate segmentation module 52, character recognition module 53 and color recognition module 54, and license plate recognition module 5 establishes data connection with user interaction module 6, license plate positioning module 51 is used for license plate positioning, license plate segmentation module 52 is used to segment the located license plate area from the image, character recognition module 53 is used to recognize the character information on the license plate, and color recognition module 54 is used to recognize the color information of the license plate to assist in the verification of the license plate; license plate positioning module 51 includes residual block module 511, attention block ... Mechanism module 512, anchor frame optimization module 513 and positioning algorithm module 514, residual block module 511 is used to optimize the extraction capability of license plate features, enhance the network's ability to capture deep-level features, and effectively alleviate the gradient vanishing problem through jump connections, so that the network can still maintain high performance while increasing depth. Attention mechanism module 512 is used for the neural network to autonomously adjust the importance weights of each part of the input data, enhance the weights of information that is crucial to license plate recognition, and weaken other unnecessary parts, thereby guiding the network to focus more on the license plate area and significantly improve the feature extraction network's attention to the license plate. Anchor frame optimization module 513 is used to optimize the anchor frame, and positioning algorithm module 514 is used to locate the license plate;The attention mechanism module 512 includes a channel attention module 5121 and a spatial attention module 5122. The channel attention module 5121 can improve the performance of the model in various tasks, such as image classification, target detection, etc. by adjusting the weights of different channels. The spatial attention module 5122 is used to dynamically suppress irrelevant information in the input data and concentrate on processing important parts, thereby improving the robustness and generalization ability of the model; the video acquisition module 1 includes a high-position camera module 11, a video encoding module 12 and a network communication module 13, and the network communication module 13 establishes a data connection with the image preprocessing module 2. The high-position camera module 11 is used to capture video images in the parking lot, and then the captured video images are encoded and processed through the video encoding module 12 for subsequent analysis and recognition, and the video data is transmitted to the image preprocessing module 2 through the network communication module 13 for image preprocessing; the image preprocessing module 2 includes an image denoising module 21, an image enhancement module 22 and an image scaling module 23. The image denoising module 21 is used to denoise the captured video images. De-noising is performed to improve the image quality. The image enhancement module 22 is used to enhance the image to make the image information clearer and more specific. The image scaling module 23 is used to scale the image as needed to meet different analysis requirements. The user interaction module 6 includes a user interface module 61, a permission management module 62 and a user feedback module 63. The user interface module 61 is used to provide a user interface to facilitate user operation. The permission management module 62 is used to manage user permissions to ensure the security of the system. The user feedback module 63 is used to collect user feedback to optimize and improve the system. The data storage management module 7 includes a database module 71, a data backup module 72, a data query module 73 and a data transmission module 74. The data transmission module 74 establishes a data connection with the user interaction module 6. The database module 71 is used to store data such as vehicles, parking spaces, and behaviors in the parking lot. The data backup module 72 is used to regularly back up the data in the database to prevent data loss. The data query module 73 is used by users to query the data in the database. The data transmission module 74 is used to export or import the data in the database.

[0027] Working principle: When the present invention is applied to parking lot video recognition management, firstly, the high-position camera module 11 in the video acquisition module 1 is used to capture the video image in the parking lot, and then the captured video image is encoded and processed by the video encoding module 12 to facilitate subsequent analysis and recognition, and the video data is transmitted to the image preprocessing module 2 through the network communication module 13 for image preprocessing, and the captured video image is denoised by the image denoising module 21 to improve the image quality, and the image enhancement module 22 enhances the image to make the image information clearer and more specific, and the image scaling module 23 is used to scale the image as needed to meet different analysis requirements, and the processed image data is transmitted to the behavior analysis module 3 for To detect abnormal behavior of pedestrians, first, the target detection module 31 outputs the pedestrian position information through the YOLOv4 algorithm, then divides the parking lot surveillance video into video frames, uses the non-maximum suppression algorithm to remove overlapping frames, further screens the candidate frames, and uses the Kalman filter to calculate the position and state of the pedestrian target in the next frame. The pedestrian detection frame with the highest confidence is used as the prediction result, and the distance between the pedestrian detection frame and the pedestrian tracking frame is calculated. When the distance between the two is less than a certain threshold, the two are correlated with each other, and the matching successfully completes the pedestrian target detection. Then, the target action trajectory is tracked by the target tracking module 32, and the pedestrian motion trajectory features are extracted by the trajectory feature extraction module 33, and the skeleton features are extracted by the skeleton feature extraction module 34. By introducing the symmetric space The inter-transformation network accurately extracts high-quality human body areas, and adjusts the posture distance parameters through the parameterized posture non-maximum suppression method to filter out erroneous detection frames and redundant information. Then, the skeleton key point data is reinforced and trained to obtain skeleton features. The trajectory features and skeleton features are fused through the feature fusion module 35. Then, based on the defined abnormal behaviors, the pedestrian abnormality analysis module 36 performs abnormal behavior analysis. The vehicle abnormality analysis module 37 is used to identify abnormal behaviors of vehicles, such as going against the flow and speeding. At the same time, the parking space detection module 4 performs parking space related detection through video data. The occupancy detection module 41 is used to detect whether the parking space is occupied. The parking space status update module 42 is used to update the status information of the parking space in real time, such as idle, occupied, etc. The parking space allocation module 43 is used to allocate a suitable parking space for the vehicle according to the parking space situation in the parking lot. The license plate recognition module 5 is used to accurately identify the license plate to match the vehicle information. First, the license plate area is located by the license plate positioning module 51. The residual block module 511 optimizes the ability to extract license plate features and enhances the network's ability to capture deep features. The gradient disappearance problem is effectively alleviated through jump connections, so that the network can still maintain high performance while increasing its depth. The attention mechanism module 512 includes a channel attention module 5121 and a spatial attention module 5122, which are used for the neural network to autonomously adjust the importance weights of each part of the input data, enhance the information weights that are critical to license plate recognition, and weaken other unnecessary parts.Thereby guiding the network to focus more on the license plate area, significantly improving the attention of the feature extraction network to the license plate, and then optimizing the anchor frame through the anchor frame optimization module 513, locating the license plate through the positioning algorithm module 514, and combining the color recognition module 54 to identify the color information of the license plate to assist in the verification of the license plate, the license plate segmentation module 52 segments the located license plate area from the image and identifies the character information on the license plate through the character recognition module 53 to achieve accurate recognition of the license plate, the user interface module 61 in the user interaction module 6 of the system is used to provide a user interface to facilitate user operation, the authority management module 62 is used to manage user authority to ensure the security of the system, the user feedback module 63 is used to collect user feedback to optimize and improve the system, the database module 71 in the data storage management module 7 is used to store data such as vehicles, parking spaces, and behaviors in the parking lot, the data backup module 72 is used to regularly back up the data in the database to prevent data loss, the data query module 73 is used for users to query the data in the database, and the data transmission module 74 is used to export or import the data in the database. ,

[0028] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A parking lot high-position video recognition system based on depth recognition, comprising a video acquisition module (1), characterized in that: The video acquisition module (1) is data-connected to an image preprocessing module (2), the image preprocessing module (2) is data-connected to a behavior analysis module (3), a parking space detection module (4) and a license plate recognition module (5), the behavior analysis module (3) is data-connected to a user interaction module (6), and the user interaction module (6) is data-connected to a data storage management module (7).

2. According to claim 1, a parking lot high-position video recognition system based on depth recognition is characterized in that: The behavior analysis module (3) comprises a target detection module (31), a target tracking module (32), a trajectory feature extraction module (33), a skeleton feature extraction module (34), a feature fusion module (35), a pedestrian anomaly analysis module (36) and a vehicle anomaly analysis module (37).

3. According to claim 1, a parking lot high-position video recognition system based on depth recognition is characterized in that: The parking space detection module (4) comprises an occupancy detection module (41), a parking space status update module (42) and a parking space allocation module (43), and the parking space detection module (4) establishes a data connection with the user interaction module (6).

4. The parking lot high-position video recognition system based on depth recognition according to claim 1 is characterized in that: The license plate recognition module (5) comprises a license plate positioning module (51), a license plate segmentation module (52), a character recognition module (53) and a color recognition module (54), and the license plate recognition module (5) establishes a data connection with the user interaction module (6).

5. The parking lot high-position video recognition system based on depth recognition according to claim 4 is characterized in that: The license plate positioning module (51) includes a residual block module (511), an attention mechanism module (512), an anchor frame optimization module (513) and a positioning algorithm module (514).

6. The parking lot high-position video recognition system based on depth recognition according to claim 5 is characterized in that: The attention mechanism module (512) includes a channel attention module (5121) and a spatial attention module (5122).

7. The parking lot high-position video recognition system based on depth recognition according to claim 1 is characterized in that: The video acquisition module (1) comprises a high-position camera module (11), a video encoding module (12) and a network communication module (13), and the network communication module (13) establishes a data connection with the image preprocessing module (2).

8. The parking lot high-position video recognition system based on depth recognition according to claim 1 is characterized by: The image preprocessing module (2) comprises an image denoising module (21), an image enhancement module (22) and an image scaling module (23).

9. The parking lot high-position video recognition system based on depth recognition according to claim 1 is characterized in that: The user interaction module (6) comprises a user interface module (61), a rights management module (62) and a user feedback module (63).

10. The parking lot high-position video recognition system based on depth recognition according to claim 1, characterized in that: The data storage management module (7) comprises a database module (71), a data backup module (72), a data query module (73) and a data transmission module (74), and the data transmission module (74) establishes a data connection with the user interaction module (6).

Images