Parking detection system and method thereof

Abstract

The present invention primarily utilizes a reading module to read a video segment and a real-time frame from a video stream according to a storage rule to form a reading segment, and an analysis module analyzes the regions of interest in at least two real-time frames and confirms the similarity between the two regions of interest. When the similarity is greater than a set threshold, determining that a parking event has occurred, and an analysis program is executed to analyze at least one reading segment and obtain a parking information of a vehicle in the reading segment. The transmission module transmits the parking information that meets the parking event to the processing device. Accordingly, the present invention can quickly determine whether the parking area has changed, and thus improve computing efficiency.

Description

Technical Field

[0001] This invention relates to parking detection technology, and more particularly to a parking detection system and method for quickly determining changes in parking space areas. Prior Technology

[0002] Modern smart parking management technology is a technical solution that uses high-position cameras and artificial intelligence technology to achieve automatic detection of parking spaces, vehicle recognition and management.

[0003] High-position cameras are installed at high locations in parking lots to cover multiple parking spaces. Image acquisition and analysis technology uses real-time images of parking spaces captured by the cameras and artificial intelligence (AI) algorithms to identify the status of parking spaces and vehicle information, then uploads and processes the information accordingly.

[0004] However, to reduce the number of devices required, high-resolution HD cameras are often needed to cover multiple parking spaces. This presents the following problems: 1. HD cameras place higher demands on the performance of the computer host terminal, resulting in high computing costs; 2. Edge computing hosts have limited decoding capabilities for HD video, making it difficult to process multiple video streams simultaneously; 3. Tracking algorithms consume significant computing power, making long-term tracking difficult and struggling to handle issues such as occlusion, slow parking, and repeated entry and exit; 4. Insufficient computing power prevents the use of high-end AI models and hinders the addition of new models and functionalities.

[0005] In view of this, based on the above problems, there is an urgent need for a technology that can quickly determine whether there are changes in parking space areas, thereby improving computing efficiency and addressing the problems of previous technologies. Summary of the Invention

[0006] In view of the aforementioned problems, the purpose of this invention is to quickly determine whether there are changes in the parking space area and improve computing efficiency, so as to solve the problems faced in the prior art.

[0007] To achieve the aforementioned objectives, the present invention provides a parking detection system, comprising: a reading module that reads a video segment and a real-time frame from an image stream according to a storage rule to form a reading segment; an analysis module connected to the reading module, the analysis module analyzing a region of interest in at least two real-time frames and determining a similarity between the two regions of interest, and when the similarity is greater than a set threshold, the analysis module determining that a parking event has occurred and executing an analysis program, the analysis module analyzing at least one reading segment according to the analysis program and obtaining parking information of a vehicle in the reading segment, and stopping the execution of the analysis program when it is determined that the vehicle has completed the parking event; and a transmission module connected to the analysis module and transmitting the parking information that matches the parking event to a processing device.

[0008] Preferably, when the reading module reads a video segment and a real-time frame from a video stream according to a storage rule to form a reading segment, the reading module reads the video segment according to a time length set by the storage rule and reads the real-time frame at a time point set by the storage rule to form the reading segment. The time point includes reading the real-time frame after the video segment ends, reading the real-time frame during the recording of the video segment, or reading the real-time frame before the start of the video segment.

[0009] Preferably, when the analysis module analyzes a region of interest in at least two real-time frames and confirms a similarity between the two regions of interest, the analysis module extracts a plurality of feature points in the two regions of interest and calculates these feature points according to a cosine similarity algorithm to confirm the similarity.

[0010] Preferably, when the analysis module executes the analysis program, the analysis module detects the vehicle's position and identifies vehicle information based on a plurality of image frames and the real-time frame in the video segment. The analysis module analyzes the image frames and the real-time frame frame by frame to obtain a vehicle detection box for the vehicle, and determines the overlap area between the region of interest and the vehicle detection box based on an overlap threshold. When the overlap area exceeds the overlap threshold, the analysis module obtains the parking information of the vehicle in the read segment and determines that the vehicle has completed the parking event.

[0011] Preferably, when the analysis module executes the analysis procedure, the analysis module detects the position of the vehicle and identifies the vehicle according to the two real-time frames, and judges an overlapping area between the region of interest and the vehicle based on an overlap threshold. When the overlapping area exceeds the overlap threshold, the analysis module acquires the parking information of the vehicle in the read segment and determines that the vehicle has completed the parking event.

[0012] Preferably, the parking detection system comprises: a storage module connected to the reading module and the analysis module, wherein the storage module acquires and stores the read segments from the reading module, and sends at least one read segment to be analyzed by the analysis module during execution of the analysis procedure to the analysis module, and the storage module deletes part of the read segments based on a deletion rule, wherein the deletion rule comprises deleting a plurality of read segments exceeding a set number, deleting a plurality of read segments exceeding a set validity period, or a combination of the two.

[0013] Preferably, when the reading module reads a plurality of the image streams, the parking detection system comprises: a load controller connected to the reading module, the storage module and the analysis module, wherein the load controller checks the number of unanalyzed read segments in each of the image streams in the storage module at a detection time, and marks the image stream with the largest number of unanalyzed read segments as Q(cqmax), the load controller performs load control according to the following rules: when Q(cqmax)<Qth, the load controller does not perform load control; when Q(cqmax)>Qth, the load controller executes a first-level load control; when Qsum > N × Qth, the load controller further enters a second-level load control on the basis of executing the first-level load control; wherein, Qth represents a load threshold for the number of the read segments, Qsum represents the total number of unanalyzed read segments in each of the image streams, and N represents the number of the image streams.

[0014] Preferably, when the load controller performs a first-level load control, the load controller causes the analysis module to preferentially analyze the read segments of the video stream corresponding to Q(cqmax); when the load controller performs a second-level load control, the load controller causes the read module to modify the storage rule according to the following rule to increase the duration of the recorded segment of the video stream corresponding to Q(cqmax): every interval TV(cqmax), the load controller checks the video stream corresponding to Q(cqmax), and when Q(cqmax)>Qth, the load controller causes the read module to modify the storage rule according to TV(cqmax)+S, wherein S is 0.25 × TV0(cqmax), and when TV(cqmax) reaches 2 × TV0(cqmax), the duration of TV(cqmax) is no longer increased; wherein, TV(cqmax) represents the currently set duration of the recorded segment, and TV0(cqmax) represents the originally initially set duration of the recorded segment.

[0015] Preferably, when the load controller checks the storage module at the detection time and performs the second-level load control, the load controller determines that Q(cqmax)<Qth, then the load controller does not perform load control, and restores TV(cqmax) to TV0(cqmax).

[0016] To achieve the above object, the present invention also provides a parking detection method applied to the parking detection system as described above, comprising: using a read module to read a recorded segment and a real-time frame from a video stream according to a storage rule to form a read segment; using an analysis module to analyze a region of interest in at least two of the real-time frames, and determining a similarity between the two regions of interest; when the similarity is greater than a set threshold, the analysis module determines that a parking event has occurred to execute an analysis procedure; analyzing at least one of the read segments according to the analysis procedure to obtain parking information of a vehicle in the read segment, and stopping executing the analysis procedure when it is determined that the vehicle completes the parking event; and using a transmission module to transmit the parking information conforming to the parking event to a processing device.

[0017] To make the above objects, features and advantages of the present invention more obvious and understandable, specific embodiments illustrated in the accompanying drawings are described below in detail. Brief Description of the Drawings

[0018] Figure 1 is a schematic diagram of the configuration relationship of the parking detection system of the present invention; Figure 2 is a schematic diagram of the composition of the read segment of the present invention; Figure 3 is a schematic diagram illustrating the steps of assembling a reading segment using the reading module of the present invention; Figures 4A and 4B are schematic diagrams showing the overlap area between the region of interest and the vehicle detection frame of the present invention; Figure 5 is a schematic diagram of the steps for analyzing and reading segments in this invention. Figure 6 is a schematic diagram of the configuration relationship of the storage module of the present invention; Figure 7 is a schematic diagram of the configuration relationship of the load controller of the present invention; Figure 8 is a flowchart of the parking detection method of the present invention. Implementation

[0019] The advantages, features, and technical methods of this invention will be more readily understood by referring to the exemplary embodiments and accompanying drawings. This invention may be implemented in different forms and should not be construed as being limited to the embodiments set forth herein. Rather, the embodiments provided will make this disclosure more thorough, complete, and fully convey the scope of the invention to those skilled in the art, and the invention will be defined only as provided in the appended claims.

[0020] Additionally, the terms "comprising" and / or "including" refer to the presence of the stated features, regions, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, regions, wholes, steps, operations, elements, components, and / or combinations thereof.

[0021] To facilitate your understanding of the content of this invention and its effects, the following detailed description is provided in conjunction with the accompanying drawings of various specific embodiments:

[0022] Please refer to Figures 1 to 5, which are schematic diagrams of the configuration relationship of the parking detection system of the present invention, schematic diagrams of the composition of the reading segment, schematic diagrams of the steps of the reading module composing the reading segment, schematic diagrams of the steps of analyzing the reading segment, and schematic diagrams of the overlap area between the region of interest and the vehicle detection frame. As shown in the figures, the present invention mainly discloses a parking detection system, which includes a reading module 10, an analysis module 20, and a transmission module 30. The reading module 10 is a related reader for reading at least one video stream 40 (the video stream 40 may originate from various optical image sensors). When reading the video stream 40 (and of course, the reading module 10 can also read multiple video streams 40), the reading module 10 reads a video segment and a real-time frame from the video stream 40 according to a storage rule 11 and combines them into a reading segment 12. The storage rule 11 may be set to read the video segment for a certain time length and read the real-time frame at a time point set by the storage rule 11 to form the reading segment 12. The time point includes reading the real-time frame after the video segment ends, reading the real-time frame during the recording of the video segment, or reading the real-time frame before the start of the video segment. In this embodiment, the real-time frame is read after the recording segment ends. That is, the real-time frame is obtained by the reading module 10 immediately from the video stream when the recording segment stops after the duration meets the specified time length.

[0023] Please refer to Figures 2 and 3. When the reading module 10 reads the c-th video stream 40 and assembles the reading segment 12, the reading module 10 sets the storage time length of the recorded video segment to TV(c), and the number of unanalyzed reading segments 12 is Q(c). TVS(c,t) and TVE(c,t) are the start and end times of the recorded video segment V(c,t) corresponding to the t-th reading segment 12 R(c,t) of the c-th video stream 40, respectively. F(c,t) is the first real-time frame obtained from the real-time video stream 40 after the recording segment V(c,t) ends, and its corresponding time point is TF(c,t). Recording and storage begin at time TVS(c,t), and recording stops when the duration of the recorded video segment reaches TV(c). The real-time frame F(c,t) is immediately obtained from the real-time video stream 40 and stored. The number of unanalyzed reading segments 12, Q(c), is... After Q(c)+1, the next process begins.

[0024] The analysis module 20 is a processor that determines whether to execute an analysis program based on the acquired plurality of read segments 12. When the analysis module 20 is connected to the read module 10 to acquire the read segments 12, the analysis module 20 can determine the similarity between two regions of interest IA defined in at least two real-time frames. The region of interest IA is a designated parking area (wherein the defined region of interest IA can be pre-determined by the analysis module or other related modules through full-image detection to draw and designate the parking area as the region of interest IA, or the designated parking area can be pre-labeled and the rectangular coordinates of the designated parking area can be stored as the region of interest IA). The similarity is detected based on the frames before and after the designated parking area. When the similarity is greater than a set threshold, the analysis module 20 determines that a parking event has occurred and executes the analysis program. When the analysis module 20 confirms the similarity between two regions of interest (ROIs) IA, it can extract multiple feature points from the two ROIs IAs using image hash algorithms such as pHash, aHash, and dHash (or extract feature points from the two ROIs IAs using a res18 neural network on the real-time frame in RGB format), and calculate these feature points using a cosine similarity algorithm to confirm the similarity. Thus, the similarity detection only requires processing two images (i.e., the two real-time frames), and the computational load is significantly less than that of frame-by-frame processing using video images.

[0025] When the analysis module 20 executes the analysis program, it can decode the video segment using a video decoding module to obtain multiple image frames in RGB or BGR format. Then, based on these image frames and the real-time frame, it detects the vehicle's position and identifies vehicle information 21 (e.g., license plate information, vehicle type, vehicle color, and vehicle orientation). When the analysis module 20 analyzes these image frames and the real-time frame by frame, it can utilize a tracking matching algorithm (e.g., using FairMOT (Fair Multi-Object Array)). The tracking method matches vehicles in consecutive frames and obtains their driving trajectories. Based on these trajectories, a vehicle detection box (VDF) is obtained. An overlap area (OA) between the region of interest (Io) and the VDF is determined according to an overlap threshold. When the overlap area (OA) exceeds the overlap threshold, the analysis module 20 determines that the vehicle has completed the parking event. Subsequently, the analysis module 20 obtains parking information 22 (e.g., license plate information, vehicle information, vehicle station status) of the vehicle from the read segment 12 and stops executing the analysis program when it is determined that the vehicle has completed the parking event.

[0026] For example, the overlap threshold can be 0% or 90%. When the analysis module 20 determines the overlap area OA between the region of interest IA and the vehicle detection frame VDF based on an overlap threshold, it can determine whether the parking event is completed based on whether the overlap area OA decreases to 0% or increases to 90%. For the vehicle within the designated parking area, if the overlap area OA between the vehicle detection frame VDF and the region of interest IA decreases until the vehicle leaves the region of interest IA (that is, the overlap area OA between the vehicle detection frame VDF and the region of interest IA decreases to 0%), the analysis module 20 confirms that the vehicle has left the designated parking area to complete the parking event. Please refer to Figure 4A. For the vehicle outside the designated parking area, if the overlap area OA between the vehicle detection frame VDF and the region of interest IA increases and reaches 90% of the overlap threshold, the analysis module 20 confirms that the vehicle has entered the designated parking area to complete the parking event. Please refer to Figure 4B.

[0027] The transmission module 30 is a transmitter for inputting or outputting relevant information. When the transmission module 30 is connected to the analysis module 20, the transmission module 30 can transmit the parking information 22 that matches the parking event to a processing device 50 in a prescribed data format, so that the processing device 50 can perform subsequent processing procedures.

[0028] Therefore, referring to Figure 5, when the analysis module 20 analyzes the read segment 12, the following steps can be followed:

[0029] The analysis module analyzes the read segment of the image stream from channel c;

[0030] Determine if the number of unanalyzed read segments Q(c) is greater than 0;

[0031] When Q(c) is greater than 0, the analysis module analyzes and reads the segment;

[0032] Determine whether the similarity between the two real-time frames is greater than a set threshold;

[0033] When the similarity is greater than a set threshold, it is determined that the scene has not changed, and the read segment is marked as an analyzed read segment. The image stream analysis progress is TF(c,t), and the number of unanalyzed read segments is Q(c) = Q(c)-1;

[0034] If the similarity does not exceed the set threshold, it is determined that the scene has changed, and the analysis module executes the analysis program.

[0035] Determine whether the parking event has been completed;

[0036] When a parking event has been completed, retrieve the parking information of the vehicle from the read segment;

[0037] Send parking information;

[0038] If a parking event is not completed, the read segment is marked as an analyzed read segment, the image stream analysis progress is TF(c,t), and the number of unanalyzed read segments is Q(c) = Q(c)-1;

[0039] After marking the read segment as analyzed, the analysis phase for the next read segment begins (t=t+1).

[0040] When Q(c) is not greater than 0, the analysis module waits for w seconds before determining whether the number of unanalyzed read segments Q(c) is greater than 0.

[0041] Additionally, when the analysis module 20 executes the analysis program, it can also detect the vehicle's position and identify the vehicle based solely on the two real-time frames, and determine the overlap area OA between the region of interest IA and the vehicle according to the overlap threshold. That is, if the designated parking area in the current real-time frame does not overlap with the vehicle, but the designated parking area in the subsequent real-time frame does overlap with the vehicle, the analysis module 20 can determine the overlap area OA according to the overlap threshold. Alternatively, if the designated parking area in the current real-time frame does overlap with the vehicle, but the designated parking area in the subsequent real-time frame does not overlap with the vehicle (or the overlap area OA decreases), the analysis module 20 can also determine the overlap area OA according to the overlap threshold. When the overlap area OA exceeds the overlap threshold (e.g., 80% and 5%, which can be adjusted according to different analysis methods), the analysis module 20 obtains the vehicle's parking information 22 from the read segment 12 and determines that the vehicle has completed the parking event.

[0042] Furthermore, the parking detection system of the present invention can also be configured to trace RN read segments 12 to achieve long-term analysis of complex events.

[0043] Please refer to Figure 6, which is a schematic diagram of the configuration relationship of the storage module of the present invention. As shown in the figure, the parking detection system of the present invention may include a storage module 60, which may be a memory used to store relevant information and provide analysis for relevant modules. The storage module 60 may be configured within the reading module 10 or be an independent module, which is not limited here. When the storage module 60 is connected to the reading module 10 and the analysis module 20, the storage module 60 can obtain and store the reading segment 12 from the reading module 10. When the analysis module 20 needs to analyze the reading segment 12, the storage module 60 can send at least one reading segment 12 that the analysis module 20 needs to analyze to perform the analysis program, so as to facilitate the analysis module 20 to judge the subsequent two real-time frames or execute the analysis program, etc.

[0044] Furthermore, the storage module 60 can delete portions of the read segments 12 based on a deletion rule 61. This deletion rule 61 includes deleting a plurality of read segments 12 exceeding a set number, deleting a plurality of read segments 12 exceeding a set expiration date, or a combination of both. Deleting a plurality of read segments 12 exceeding a set number indicates that the user can set a maximum storage limit. When the stored read segments 12 exceed this set limit, the storage module 60 can delete the oldest and most analyzed read segments 12. Deleting a plurality of read segments 12 exceeding a set expiration date indicates that the user can set a storage validity period. When the stored read segments 12 exceed this set expiration date (e.g., 3 months or 6 months), the storage module 60 can delete the read segments 12 that have exceeded the set expiration date and have already been analyzed.

[0045] When the storage module 60 obtains and stores the read segment 12 from the read module 10, the storage module 60 can store the video segment in the read segment 12 as MP4 format and record the start and end times of the video segment, while the real-time frame in the read segment 12 is directly stored as BGR format.

[0046] Please refer to Figure 7, which is a schematic diagram of the configuration relationship of the load controller of the present invention. As shown in the figure, when the reading module 10 reads a plurality of the image streams 40, the parking detection system of the present invention may include a load controller 70, which is used to adjust the priority order of the analysis of each image stream 40 in the analysis module 20 according to the analysis progress of each image stream 40, and adjust the parameters of the reading module 10 to change the speed at which each image stream 40 forms the reading segment 12, thereby balancing the analysis progress of each image stream 40.

[0047] When the load controller 70 is connected to the reading module 10, the storage module 60 and the analysis module 20, the load controller 70 may check the quantity Q of the unanalyzed read segments 12 in each of the image streams 40 in the storage module 60 at a detection time (for example, at regular intervals of delta seconds), and mark the image stream 40 having the largest number of unanalyzed read segments 12 as Q(cqmax). The load controller 70 performs load control according to the following rules:

[0048] Q(cqmax)<Qth: the load controller does not perform load control;

[0049] Q(cqmax)>Qth: the load controller executes a first-level load control;

[0050] Qsum > N × Qth: the load controller further enters a second-level load control under the condition of executing the first-level load control;

[0051] wherein Qth represents a load threshold of the number of read segments, Qsum represents the total number of unanalyzed read segments in each image stream, and N represents the number of the image streams.

[0052] When the load controller 70 does not perform load control, it means that the load controller 70 continues to check the storage module 60 but does not perform intervention control actions.

[0053] When the load controller 70 executes first-level load control, the load controller 70 drives the analysis module 20 to preferentially analyze the read segments 12 of the image stream 40 corresponding to Q(cqmax), so as to preferentially process the backlogged image stream 40 to balance the analysis delay of each image stream 40, and the analysis process of the remaining read segments 12 enters a sleep state first until the next load check is performed.

[0054] When the load controller 70 executes second-level load control, it mainly acts on the reading module 10, and reduces the number of read segments 12 to be analyzed per unit time by increasing the time interval of the recorded video segments, so the load controller 70 can drive the reading module 10 to modify the storage rule 11 according to the following rules, so as to increase the time interval of the recorded video segments corresponding to Q(cqmax):

[0055] Every interval TV(cqmax), the load controller 70 checks the video stream 40 corresponding to Q(cqmax). When Q(cqmax)>Qth, the reading module 10 is driven to modify the storage rule 11 according to TV(cqmax)+S, where S is 0.25 × TV0(cqmax). When TV(cqmax) reaches 2 × TV0(cqmax), the time length of TV(cqmax) will not be further increased; wherein, TV(cqmax) represents the currently set time length of the recorded video segment, and TV0(cqmax) represents the originally initially set time length of the recorded video segment.

[0056] In addition, when the load controller 70 checks the storage module 60 at the detection time and executes the second-level load control, the load controller 70 determines that Q(cqmax)<Qth, then the load controller 70 does not perform load control, and restores TV(cqmax) to TV0(cqmax).

[0057] Thereby, the load controller 70 can realize delay balance for multi-channel video stream 40 analysis.

[0058] Furthermore, the present application can also add a third-level load control, which does not perform the tracking analysis of the analysis module as described above, and only judges the parking information 22 of the vehicle based on the information of each real-time frame.

[0059] Please refer to Fig. 8 again, which is a step flow chart of the parking detection method of the present invention. As shown in the figure, the present invention provides a parking detection method applicable to the parking detection system as described above, which comprises the following steps:

[0060] S101: using a reading module to read recorded video segments and real-time frames from a video stream according to a storage rule to form a reading segment;

[0061] S102: using an analysis module to analyze a region of interest in at least two of the real-time frames, and confirm the similarity between the two regions of interest;

[0062] S103: when the similarity is greater than a set threshold, the analysis module determines that a parking event has occurred to execute an analysis procedure;

[0063] S104: analyzing at least the read segment according to the analysis procedure and acquiring parking information of the vehicle in the read segment, and stopping executing the analysis procedure when it is determined that the vehicle has completed the parking event; and

[0064] S105: The parking information that matches the parking event is transmitted to the processing device using the transmission module.

[0065] The embodiments disclosed in this invention are preferred embodiments. Any local changes or modifications that originate from the technical concept of this invention and can be easily deduced by those skilled in the art are not outside the scope of the patent rights of this invention.

[0066] In conclusion, this invention, in terms of its purpose, means, and effects, demonstrates technical features that are distinct from those of the conventional invention. Furthermore, its invention is practical and meets all the requirements for a patent. We earnestly request that your esteemed examiner carefully review this matter and grant a patent as soon as possible so that it may benefit society. We would be truly grateful for your assistance.

[0067] 10: Read Module 11: Storage Rules 12: Reading a fragment 20: Analysis Module 21: Vehicle Information 22: Parking Information 30: Transmission Module 40: Video Stream 50: Processing device 60: Storage Module 61: Deletion Rules 70: Load Controller IA: Region of Interest VDF: Vehicle Inspection Frame OA: Overlapping area S101-S105: Steps

Claims

1. A parking detection system, comprising: A reading module reads a video segment and a real-time frame from an image stream according to a storage rule to form a reading segment; an analysis module connected to the reading module analyzes a region of interest in at least two real-time frames and determines a similarity between the two regions of interest. When the similarity is greater than a set threshold, the analysis module determines that a parking event has occurred and executes an analysis program. The analysis module analyzes at least one reading segment according to the analysis program and obtains parking information of a vehicle in the reading segment. The analysis program is stopped when it is determined that the vehicle has completed the parking event; a transmission module connected to the analysis module transmits the parking information that matches the parking event to a processing device. A storage module connected to the reading module and the analysis module, the storage module acquires and stores the read segment from the reading module, and sends at least one read segment to the analysis module for analysis of the analysis program. The storage module deletes portions of the read segment based on a deletion rule, wherein the deletion rule includes deleting a plurality of read segments exceeding a set number, deleting a plurality of read segments exceeding a set expiration period, or a combination of both. A load controller connected to the reading module, the storage module, and the analysis module, when the reading module reads a plurality of image streams, the load controller checks the number of unanalyzed read segments in each image stream in the storage module at a detection time, and marks the image stream with the largest number of unanalyzed read segments as Q(cqmax). The load controller performs load control according to the following rule: if Q(cqmax) < Qth, the load controller does not perform load control. If Q(cqmax) > Qth, the load controller performs a first-level load control; if Qsum > N × Qth, the load controller, while performing the first-level load control, then enters a second-level load control; where Qth represents a load threshold for the number of read segments, Qsum represents the total number of unanalyzed read segments in each image stream, and N represents the number of image streams.

2. The parking detection system as described in claim 1, wherein, When the reading module reads a video segment and a real-time frame from a video stream according to a storage rule to form a reading segment, the reading module reads the video segment according to a time length set by the storage rule and reads the real-time frame at a time point set by the storage rule to form the reading segment. The time point includes reading the real-time frame after the video segment ends, reading the real-time frame during the recording of the video segment, or reading the real-time frame before the start of the video segment.

3. The parking detection system as described in claim 1, wherein, When the analysis module analyzes a region of interest in at least two real-time frames and confirms a similarity between the two regions of interest, the analysis module extracts a plurality of feature points from the two regions of interest and calculates these feature points according to a cosine similarity algorithm to confirm the similarity.

4. The parking detection system as described in claim 1, wherein, When the analysis module executes the analysis program, it detects the vehicle's position and identifies vehicle information based on multiple image frames and real-time frames in the video clip. The analysis module analyzes the image frames and real-time frames frame by frame to obtain a vehicle detection box. It then determines the overlap area between the region of interest and the vehicle detection box based on an overlap threshold. When the overlap area exceeds the overlap threshold, the analysis module obtains the vehicle's parking information in the read segment and determines that the vehicle has completed the parking event.

5. The parking detection system as described in claim 1, wherein, When the analysis module executes the analysis program, it detects the vehicle's position and identifies the vehicle based on the two real-time frames, and determines the overlap area between the region of interest and the vehicle based on an overlap threshold. When the overlap area exceeds the overlap threshold, the analysis module obtains the vehicle's parking information in the read segment and determines that the vehicle has completed the parking event.

6. The parking detection system as described in claim 1, wherein, When the load controller performs a first-level load control, it instructs the analysis module to prioritize analyzing the read segments of the video stream corresponding to Q(cqmax). When the load controller performs a second-level load control, it instructs the read module to modify the storage rule according to the following rule to increase the duration of the video segment corresponding to Q(cqmax): Every TV(cqmax), the load controller checks the video stream corresponding to Q(cqmax). When Q(cqmax) > Qth, it instructs the read module to modify the storage rule according to TV(cqmax) + S, where S represents 0.25 × TV0(cqmax). When TV(cqmax) reaches 2 × TV0(cqmax), the duration of TV(cqmax) is no longer increased. Here, TV(cqmax) represents the currently set duration of the video segment, and TV0(cqmax) represents the originally set duration of the video segment.

7. The parking detection system as described in claim 6, wherein, When the load controller checks the storage module and performs the second-level load control at the detection time, if the load controller determines that Q(cqmax) < Qth, the load controller will not perform load control and will restore TV(cqmax) to TV0(cqmax).

8. A parking detection method applied to a parking detection system as described in any one of claims 1 to 7, comprising: A reading module is used to read a video segment and a real-time frame from a video stream according to a storage rule to form a reading segment; An analysis module is used to analyze a region of interest in at least two real-time frames and to determine a similarity between the two regions of interest. When the similarity is greater than a set threshold, the analysis module determines that a parking event has occurred and executes an analysis program. The analysis program is used to analyze at least one read segment and to obtain parking information of a vehicle in the read segment. The analysis program is stopped when it is determined that the vehicle has completed the parking event. And using a transmission module to transmit the parking information that matches the parking event to a processing device; The parking detection method further includes: acquiring and storing the read segment from the read module using a storage module, and sending at least one read segment required for analysis by the analysis module to perform the analysis program to the analysis module; the storage module deleting a portion of the read segment based on a deletion rule, wherein the deletion rule includes deleting a plurality of read segments exceeding a set number, deleting a plurality of read segments exceeding a set validity period, or a combination of both; and when the read module reads a plurality of the image streams, using a load controller to check the number of unanalyzed read segments in each of the image streams in the storage module at a detection time, and marking the image stream with the largest number of unanalyzed read segments as Q(cqmax); the load controller performs load control according to the following rules: Q(cqmax) < Qth, the load controller does not perform load control; Q(cqmax) > Qth, the load controller performs a first-level load control; Qsum > N × Qth, the load controller, while performing the first-level load control, then enters a second-level load control. Where Qth represents a load threshold for the number of read segments, Qsum represents the total number of unanalyzed read segments in each of the image streams, and N represents the number of the image streams.

Images