Two-wheeled vehicle battery carrying behavior safety monitoring system and method based on posture recognition

Abstract

According to the two-wheeled vehicle battery carrying behavior safety monitoring system and method based on posture recognition, the hand area of a passenger is locked through posture recognition, then the image of a carried object is extracted and analyzed in the hand area, and targeted monitoring of battery carrying disassembling behaviors such as battery lifting and battery holding is achieved. Meanwhile, the detection range is narrowed based on the attitude data, only images intercepted in the adjacent area of the hand are analyzed, irrelevant background interference is reduced, the calculation amount of image recognition is reduced, and the accuracy and real-time performance of carried object recognition are improved. When a passenger just enters the elevator and the elevator is not started, carried object detection is completed, once the passenger is recognized as an electric vehicle battery, a warning instruction is output, and an elevator control strategy can be linked, so that the safety risk of the battery can be recognized and blocked at the source of entering a closed lift car; therefore, the probability of fire disasters and casualties caused by thermal runaway of the electric vehicle battery in the elevator space is remarkably reduced, and the overall safety of the elevator using scene is improved.

Description

Technical Field

[0001] This invention relates to the technical field of safety monitoring systems for battery carrying behavior in two-wheeled vehicles based on posture recognition, and particularly to a safety monitoring system and method for battery carrying behavior in two-wheeled vehicles based on posture recognition. Background Technology

[0002] Existing electric vehicles typically consist of a braking system, a battery, a drive system, and other components. The battery supplies power to each system to drive the electric vehicle.

[0003] When a battery is depleted, there are two main ways to recharge it: First, park the vehicle at a designated charging station and charge it using a fixed charging point. Second, remove the battery from the vehicle and take it back to your residence or workplace to charge it using an external charging cable. For residents of high-rise buildings, the latter is more common; they often carry the removed battery into the elevator and then take it upstairs.

[0004] However, the lithium batteries used in electric vehicles inherently carry the risk of thermal runaway. Under abnormal conditions such as overcharging, cell defects, compression, or collisions, the batteries may experience intense heating, even catching fire or exploding within a short period. Elevator cars are enclosed spaces with limited volume and relatively dense crowds. If a battery brought into an elevator experiences thermal runaway, it will burn rapidly, release a large amount of heat, and produce a large amount of toxic fumes. Passengers will have almost no effective escape or evacuation space, which can easily lead to casualties and serious elevator safety accidents.

[0005] In existing technologies, for situations where electric vehicles are pushed into elevators as a whole, existing solutions involve placing cameras inside the elevator car and using image recognition algorithms to identify the overall outline or features of the vehicle. When an electric vehicle is detected being pushed in, a voice or text warning is issued, or elevator operation is restricted. However, these solutions primarily target the identification of electric vehicles in their entirety. For batteries that have been removed from the vehicle, due to their small size, various carrying methods, and ease of concealment by clothing or luggage, relying solely on whole-vehicle image recognition is insufficient to promptly and accurately identify the carrying of electric vehicle batteries. This leaves the safety risks of removing batteries and bringing them into the elevator unmonitored and unmanageable. Summary of the Invention

[0006] Therefore, it is necessary to propose a safety monitoring system and method for battery carrying behavior of two-wheeled vehicles based on posture recognition to solve the above technical problems.

[0007] On the one hand, a posture recognition-based safety monitoring system for battery-carrying behavior in two-wheeled vehicles is provided, the system comprising: The image acquisition module is used to acquire video image sequences during the process of passengers entering the elevator; The recognition module is used to identify the posture data of passengers in a video image sequence; The judgment module determines whether the hands of the passenger in the video image sequence are carrying items based on the posture data. If items are carried, the image area corresponding to the items is used as the detection object and a detection instruction is generated. The analysis module executes the detection command to perform image analysis on the object in the video image sequence and generate analysis data. If the object in the analysis data is a battery, a warning command is generated. In at least one embodiment of this application, the analysis module includes: The text extraction module is used to extract text data from the surface of the object and parse the text data to obtain the parsed text data. The shape pattern extraction module is used to extract the shape pattern data of the surface of the object being carried.

[0008] In at least one embodiment of this application, the analysis module further includes: The matching module performs battery pattern matching on the shape pattern data. If a match is found, pattern matching data is generated. The results output module generates analysis data based on the parsed text data and / or the pattern matching data.

[0009] On the other hand, a method for monitoring the safety of battery-carrying behavior in two-wheeled vehicles based on posture recognition is provided, applied to the posture recognition-based safety monitoring system for battery-carrying behavior in two-wheeled vehicles as described in any one of the above-mentioned methods, wherein the posture recognition-based safety monitoring method for battery-carrying behavior in two-wheeled vehicles includes: Acquire video image sequences of passengers entering the elevator; The video image sequence is parsed to generate a set of images of the passenger's hands, and the set of hand images is used to determine whether the passenger is carrying any items. If the passenger is carrying any items, a detection command is generated. The detection command is executed to analyze the objects carried in the hand image set and generate analysis data. If the item carried in the analyzed data is a battery, a warning instruction is generated.

[0010] In at least one embodiment of this application, the specific steps of executing the detection instruction, parsing the objects carried in the hand image set, and generating analysis data include: Establish a database of battery textual features; The text on the object being carried in the hand image set is analyzed to generate text data; Determine the correlation between the text data and the preset battery text feature data in the battery text feature database, and generate a text judgment probability value; The text judgment probability value is compared with a first threshold. If the text judgment probability value is greater than the first threshold, a battery confirmation result is generated, and analysis data is generated.

[0011] In at least one embodiment of this application, the specific steps of comparing the text judgment probability value with a first threshold, and generating a battery confirmation result and analysis data if the text judgment probability value is greater than the first threshold, further include: If the probability value of the text comparison is less than the first threshold, text comparison failure data is generated, and analysis data is generated based on the text comparison failure data.

[0012] In at least one embodiment of this application, the specific steps of executing the detection instruction, parsing the objects in the hand image set, and generating analysis data further include: The surface pattern on the object carried in the hand image set is analyzed to generate shape pattern data; The shape pattern data is retrieved based on a preset battery image library to obtain a set of suspected battery images. The shape pattern data is compared sequentially with each image in the suspected battery image set. If the comparison result exceeds the second threshold, a confirmed battery result is generated, and analysis data is generated.

[0013] In at least one embodiment of this application, the step of sequentially comparing the shape pattern data with each image in the suspected battery image set, and generating a confirmed battery result and analysis data if the comparison result exceeds a second threshold, further includes: If the comparison result does not exceed the second threshold, image comparison failure data is generated, and analysis data is generated based on the image comparison failure data.

[0014] In at least one embodiment of this application, the step of generating a warning instruction if the item carried in the analyzed data is a battery further includes: If the condition for confirming that the carried item is a battery is not met, a supplementary monitoring command is generated; Execute the supplementary monitoring command to obtain magnetic field change data; Based on the magnetic field change data, it is determined whether the object being carried is a battery. If it is a battery, a warning command is generated.

[0015] In at least one embodiment of this application, the specific steps of executing the supplementary monitoring command to obtain magnetic field change data of items carried by the passenger include: Acquire data on the changes in the magnetic field emitted by the excitation coil detected by the detection coil near the elevator door during the process of a passenger passing through the elevator door; Obtain the shape image of the object carried in the hand image set, and map the magnetic field change data onto the shape image to obtain the magnetic induction image of the object; Analyze the magnetic induction range in the magnetic induction image of the object and generate a magnetic induction percentage value; The magnetic induction ratio is compared with a third threshold. If it exceeds the third threshold, a warning instruction is generated.

[0016] The posture recognition-based two-wheeled vehicle battery carrying behavior safety monitoring system of this embodiment will have at least the following beneficial effects: The above-mentioned safety monitoring system and method for battery carrying behavior of two-wheeled vehicles based on posture recognition locks the hand area of ​​the passenger through posture recognition, and then extracts and analyzes the image of the carried object in the hand area. It realizes targeted monitoring of battery carrying behaviors such as carrying the battery by hand and holding the battery, and overcomes the defect of relying solely on the overall shape recognition of the vehicle, which cannot detect the removal of the battery into the elevator.

[0017] Meanwhile, by narrowing the detection range based on posture data, only images of the area adjacent to the hand are analyzed, reducing interference from irrelevant backgrounds, lowering the computational load of image recognition, and improving the accuracy and real-time performance of object recognition.

[0018] By detecting items carried by passengers as soon as they enter the elevator and before the elevator starts moving, a warning command is output once an electric vehicle battery is identified, and the elevator control strategy can be linked. This allows battery safety risks to be identified and blocked at the source of entering the enclosed car, thereby significantly reducing the probability of fire and personal injury caused by thermal runaway of electric vehicle batteries in the elevator space and improving the overall safety of elevator use scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] in: Figure 1 This is a block diagram of a two-wheeled vehicle battery-carrying behavior safety monitoring system based on posture recognition in one embodiment. Figure 2 This is a flowchart of a two-wheeled vehicle battery carrying behavior safety monitoring method based on posture recognition in one embodiment; Figure 3 This is a flowchart of a two-wheeled vehicle battery carrying behavior safety monitoring method based on posture recognition in another embodiment; Figure 4 This is a flowchart of a two-wheeled vehicle battery carrying behavior safety monitoring method based on posture recognition in another embodiment; Figure 5 This is a flowchart of a two-wheeled vehicle battery carrying behavior safety monitoring method based on posture recognition in one embodiment (using magnetic induction detection).

[0021] Explanation of main component symbols 100. Safety monitoring system for battery-carrying behavior in two-wheeled vehicles based on posture recognition; 110. Image acquisition module; 120. Recognition module; 130. Judgment module; 140. Analysis module; 150. Text extraction module; 160. Shape pattern extraction module; 170. Matching module; 180. Result output module. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] On one hand, a posture recognition-based safety monitoring system 100 for carrying batteries in two-wheeled vehicles is provided, the posture recognition-based safety monitoring system 100 for carrying batteries in two-wheeled vehicles includes: Image acquisition module 110 is used to acquire video image sequences during the process of a passenger entering the elevator; The recognition module 120 is used to recognize the posture data of a person riding in a video image sequence; The judgment module 130 determines whether the hands of the passenger in the video image sequence are carrying items based on the posture data. If items are carried, the image area corresponding to the items is used as the detection object and a detection instruction is generated. The analysis module 140 executes the detection instruction to perform image analysis on the object in the video image sequence and generate analysis data. If the object in the analysis data is a battery, a warning instruction is generated. Please refer to Figure 1 In this embodiment, the image acquisition module 110 is electrically connected to a camera installed on the top or side wall of the elevator car. During the time period when the elevator door is open or remains open, it collects video image sequences of passengers entering the elevator, ensuring coverage of the entire process of passengers entering the door and taking their positions.

[0024] The recognition module 120 performs human posture recognition on the acquired video image sequence and extracts key point data of the occupant's body skeleton, especially the joint position parameters related to the upper limbs and hands, in order to obtain the occupant's posture data.

[0025] The judgment module 130 determines the area where the person's hands are located in the video image sequence based on the posture data, and identifies whether there is a target area in the area whose outline, area, and texture are significantly different from the background. If it is determined that the hands are carrying an object, the image area corresponding to the object is extracted from the video image sequence as the detection object and a detection command is generated and sent to the analysis module 140.

[0026] After receiving the detection command, the analysis module 140 performs feature extraction and classification recognition on the image area of ​​the carried object, and outputs the recognition result of the carried object as analysis data. When the analysis data indicates that the carried object is an electric vehicle battery, the analysis module 140 generates a warning command and outputs it to the elevator control system and / or the display and voice device in the car to drive voice broadcast, prompt screen or link to restrict elevator operation, so as to achieve automatic recognition and safety warning when a passenger carries a two-wheeled vehicle battery into the elevator.

[0027] By using posture recognition to locate the hand area of ​​the passenger, and then extracting and analyzing the image of the carried items within the hand area, targeted monitoring of battery removal and carrying behaviors such as carrying batteries by hand or holding batteries is achieved, overcoming the deficiency of relying solely on vehicle shape recognition, which cannot detect battery removal and entry into elevators.

[0028] Meanwhile, by narrowing the detection range based on posture data, only images of the area adjacent to the hand are analyzed, reducing interference from irrelevant backgrounds, lowering the computational load of image recognition, and improving the accuracy and real-time performance of object recognition.

[0029] By detecting items carried by passengers as soon as they enter the elevator and before the elevator starts moving, a warning command is output once an electric vehicle battery is identified, and the elevator control strategy can be linked. This allows battery safety risks to be identified and blocked at the source of entering the enclosed car, thereby significantly reducing the probability of fire and personal injury caused by thermal runaway of electric vehicle batteries in the elevator space and improving the overall safety of elevator use scenarios.

[0030] The posture recognition-based two-wheeled vehicle battery carrying behavior safety monitoring system 100 of this embodiment is installed in an elevator car or elevator control system. It may include a camera installed on the top or side wall of the car and a processing unit electrically connected to the camera. The processing unit integrates functional units such as an image acquisition module 110, a recognition module 120, a judgment module 130 and an analysis module 140. Each module can be implemented by a combination of software programs and hardware circuits.

[0031] First, the image acquisition module 110 is used to acquire video image sequences during the process of a passenger entering the elevator. The image acquisition module 110 (camera) is connected to the elevator door operator control signal. When the elevator door opens or is in the open state, it begins to acquire video image sequences and continuously acquires multiple frames of images within the time window when the passenger enters the car, stands still, or when the elevator door is about to close, thereby covering the entire process of the passenger carrying items into and out of the elevator.

[0032] Secondly, the recognition module 120 is used to recognize the posture data of the passenger in the video image sequence. The recognition module 120 can extract the key points of the human skeleton of the passenger from the video image sequence based on the human posture estimation algorithm, including multiple key points such as the head, torso, upper limbs and lower limbs, and focus on extracting the joint positions of the left and right upper limbs, such as the shoulder, elbow and wrist.

[0033] The recognition module 120 can perform temporal analysis on the posture data in consecutive frames to determine the swing trajectory of the passenger's arm and the approximate range of the hand area, so that it can still stably lock the passenger's hand area in complex scenes.

[0034] Then, the judgment module 130 determines whether the hands of the passenger in the video image sequence are carrying items based on the posture data.

[0035] The judgment module 130 can identify whether there is a target area near the hand that is significantly different in size, shape, and texture from the background within the candidate hand area given by the recognition module 120, by combining features such as brightness changes, edge contours, motion vectors, and area in the image. When a stable target area is detected around the hand, the judgment module 130 determines that the passenger is in a carrying state.

[0036] If the judgment result is that the hand is carrying an object, the judgment module 130 further extracts the corresponding image region of the carried object from the video image sequence as the detection object for subsequent image analysis, and generates a detection command based on the detection object and passes it to the analysis module 140.

[0037] By using a two-level localization method that zooms out to the hand based on pose and then searches for the object being carried within the hand area, the computational load of performing global object detection on the entire image can be significantly reduced. This avoids misidentifying walls, button panels, billboards, etc., inside the elevator as objects to be detected, thereby reducing the false alarm rate and improving the system response speed.

[0038] Finally, the analysis module 140 executes the detection command to perform image analysis on the objects carried in the video image sequence and generate analysis data.

[0039] After receiving the image region of the object given by the judgment module 130, the analysis module 140 can extract visual features such as texture features, contour features and color distribution of the region, or perform classification and recognition of the image of the object based on a pre-trained target recognition model to obtain the recognition result that the object belongs to an electric vehicle battery, a regular backpack, a cardboard box or other items, and output the recognition result as analysis data.

[0040] The analysis module 140 can compare the confidence level corresponding to the battery category with a preset threshold. When it is determined that the object carried in the analysis data is an electric vehicle battery, the analysis module 140 generates a warning command and sends the warning command to the elevator control system and / or the car display and voice broadcast device to trigger voice prompts and screen warning information display. If necessary, it can also link the control of the elevator to keep the door open or refuse to start, thereby prompting passengers to remove the electric vehicle battery from the elevator in time to prevent high-risk batteries from entering the elevator and causing safety hazards.

[0041] By analyzing the hand state of the passenger through the posture recognition module 120 and the judgment module 130, and based on whether the hand is carrying items and the image features of the items, the transformation from vehicle shape recognition to battery carrying behavior recognition is realized, which can effectively cover high-risk scenarios of removing the battery and entering the elevator in various ways.

[0042] By accurately locating the hand area of ​​a passenger using pose data, only the image of the object being carried is cropped near the hand as the detection target, instead of performing an indiscriminate target search on the entire image. This significantly reduces the computational load of image analysis, facilitating real-time processing on elevator controllers or edge computing devices.

[0043] The image acquisition module 110 acquires video image sequences throughout the process of a passenger entering the elevator, and completes posture recognition and object recognition before the elevator starts running. Once the analysis module 140 determines that the object is an electric vehicle battery, it immediately outputs a warning command to alert the user.

[0044] It can identify and block batteries as soon as they enter the elevator, reducing the possibility of thermal runaway, fire and explosion of batteries in the elevator from the source, and significantly improving the safety level of personnel in elevator use scenarios.

[0045] In at least one embodiment of this application, the analysis module 140 includes: The text extraction module 150 is used to extract text data from the surface of the object and parse the text data to obtain the parsed text data. The shape pattern extraction module 160 is used to extract the shape pattern data of the surface of the object being carried.

[0046] In at least one embodiment of this application, the analysis module 140 further includes: The matching module 170 performs battery pattern matching on the shape pattern data. If a match is found, pattern matching data is generated. The result output module 180 generates analysis data based on the parsed text data and / or the pattern matching data.

[0047] Please refer to Figure 1 In this embodiment, the analysis module 140 includes a text extraction module 150 and a shape pattern extraction module 160. The text extraction module 150 performs character recognition on the image area corresponding to the object, extracts the text data printed or affixed to the surface of the object, such as brand name, product model, voltage level, capacity marking, safety warning, battery type marking, etc., and performs semantic parsing and structured processing on the extracted text data to obtain the parsed text data, which is convenient for subsequent matching with the preset battery text feature library.

[0048] The shape pattern extraction module 160 extracts the shape outline and pattern features of the surface of the object from the same image area, such as battery icons, safety sign graphics, manufacturer logo patterns, and strip sticker outlines, and forms shape pattern data through edge detection, feature point extraction and other methods.

[0049] The analysis module 140 also includes a matching module 170 and a result output module 180. The matching module 170 receives the shape pattern data output by the shape pattern extraction module 160, compares the shape pattern data with a pre-established battery shape pattern library, performs battery pattern matching, and generates pattern matching data when the matching result meets the preset similarity conditions, indicating that the object has battery shape features.

[0050] The result output module 180 then associates the pattern matching data output by the matching module 170 with the parsed text data output by the text extraction module 150. When the parsed text data contains key fields, parameters or category identifiers related to the electric vehicle battery, or when the pattern matching data indicates that the shape pattern is highly similar to the battery pattern, the result output module 180 generates analysis data accordingly and feeds it back to the upper-level analysis process to determine whether the object being carried is an electric vehicle battery.

[0051] By introducing a dual-channel recognition mechanism of text and pattern in the analysis module 140, the objective fact that electric vehicle battery shells are generally printed with text features such as "voltage, capacity, manufacturer information, and safety warnings" is utilized to transform the text information on the surface of the item into parsable structured data. This makes it easy to accurately identify the battery category through rules such as keywords and parameter ranges, so that the item can still be distinguished by the text content when its shape is similar to other bags and toolboxes.

[0052] On the other hand, by extracting and matching the shape patterns, representative battery icons, safety signs, brand patterns, etc. on the surface of the item are included in the identification criteria. Even if some text is difficult to identify due to dirt, obstruction, or poor shooting angle, the pattern features can still be used to provide supplementary evidence for the identity of the item. The two types of features complement each other, which improves the robustness and reliability of the identification results.

[0053] By using the output module 180 to comprehensively judge the parsed text data and pattern matching data, the system can effectively reduce the false alarm rate of misjudging ordinary luggage and toolboxes as batteries based on appearance alone without significantly increasing the complexity of the system hardware. At the same time, it can reduce the missed alarms caused by the lack of obvious single features. This allows the system to more accurately identify disassembled electric vehicle batteries under complex and varied carrying methods, thereby further improving the accuracy of safety monitoring of two-wheeled vehicle battery carrying behavior in elevator scenarios.

[0054] The analysis module 140 is integrated into the processing unit of the elevator control system. After the judgment module 130 determines that the passenger is carrying an item in their hand and generates a detection command, it performs more detailed recognition of the item image.

[0055] On the other hand, a method for monitoring the safety of battery-carrying behavior in two-wheeled vehicles based on posture recognition is provided, applied to the posture recognition-based safety monitoring system 100 for battery-carrying behavior in two-wheeled vehicles as described in any of the above-mentioned methods. The posture recognition-based safety monitoring method for battery-carrying behavior in two-wheeled vehicles includes: S101. Obtain a video image sequence of a passenger entering the elevator; S102. Parse the video image sequence to generate a set of images of the passenger's hands, and determine whether the passenger is carrying any items based on the set of hand images. If the passenger is carrying any items, generate a detection command. S103. Execute the detection instruction, parse the objects carried in the hand image set, and generate analysis data; S104. If the item carried in the analyzed data is a battery, a warning instruction is generated.

[0056] Please refer to Figure 2 In this embodiment, the camera installed inside the elevator car continuously collects video image sequences during the time when the elevator door is open and passengers are entering and exiting the elevator. The video image sequences are then input to the processing unit through the image acquisition module 110.

[0057] The processing unit first preprocesses and detects human bodies in the video image sequence, analyzes the video image sequence, extracts the human posture information of each passenger, determines the approximate location area of ​​the passenger's upper limbs and hands based on the posture recognition results, and captures hand images corresponding to multiple time points within the location area, thereby forming a set of hand images of the passenger.

[0058] Subsequently, based on the hand image set, it is analyzed whether there is a target area around the hand that is significantly different from the human skin and clothing background in terms of area, outline, and texture. If it is determined that there is a stable target area on the hand, it is determined that the passenger is carrying an item, and a detection instruction containing the location, size, and corresponding frame index information of the target area of ​​the item is generated.

[0059] When executing the detection command, the image region corresponding to the object is retrieved from the hand image set, and steps such as text extraction, shape feature extraction and / or target classification and recognition are performed on the image region. The objects in the hand image set are analyzed and analysis data containing object category information, recognition confidence and other content is generated.

[0060] When the analyzed data indicates that the carried item matches the preset characteristics of an electric vehicle battery, that is, when the carried item is determined to be a battery, the system generates a warning command that is linked to the elevator control system. This command is used to drive the voice broadcast in the car and the display screen to pop up a prompt interface. If necessary, it can also control the elevator to keep the door open or prevent it from starting, thereby enabling real-time detection and early warning when a passenger carries a two-wheeled vehicle battery into the elevator.

[0061] The detection target is located at the passenger's hands and their belongings. By analyzing the video image sequence during the elevator entry process, the system extracts a set of hand images and determines whether there are any belongings being carried at the hand image level. Then, the system performs detailed identification on the images of the belongings, enabling the system to conduct targeted monitoring of various concealed carrying methods, such as removing batteries and carrying them by hand, or putting batteries in backpacks or bags but partially exposed. This effectively solves the problem of not being able to identify the behavior of removing batteries and carrying them.

[0062] Meanwhile, by extracting images and generating detection instructions within the limited area of ​​the hand, global complex target detection of the entire video image is avoided, interference from irrelevant backgrounds is reduced, the amount of computation is reduced and the real-time performance is improved, and the risk of false alarms such as misidentifying walls, billboards, and ordinary luggage as batteries is reduced.

[0063] Because the detection of carried items and battery determination are completed as soon as the passenger enters the elevator and before the elevator starts operating, a warning instruction can be generated to alert and prevent the battery from entering the enclosed car with the passenger. This reduces the probability of electric vehicle batteries experiencing thermal runaway, fire, or explosion in the elevator from the source, significantly improving the safety of the elevator operating environment.

[0064] In at least one embodiment of this application, the specific steps of executing the detection instruction, parsing the objects carried in the hand image set, and generating analysis data include: S201. Establish a battery text feature database; S202. Analyze the text on the object carried in the hand image set to generate text data; S203. Determine the correlation between the text data and the preset battery text feature data in the battery text feature database, and generate a text judgment probability value; S204. Compare the text judgment probability value with the first threshold; S205. If the probability value of the text judgment is greater than the first threshold, then a confirmed battery result is generated, and analysis data is generated.

[0065] In at least one embodiment of this application, the specific steps of comparing the text judgment probability value with a first threshold, and generating a battery confirmation result and analysis data if the text judgment probability value is greater than the first threshold, further include: S206. If the text judgment probability value is less than the first threshold, text comparison failure data is generated, and analysis data is generated based on the text comparison failure data.

[0066] Please refer to Figure 3 In this embodiment, during the deployment phase, the system pre-collects photos of the casings and nameplate information of several mainstream two-wheeled vehicle battery products, extracts text features including brand name, series model, voltage level (such as "48V" "60V"), capacity label (such as "20Ah"), usage limitation (such as "electric vehicle only" "lithium battery pack"), and safety warnings, constructs a battery text feature database, and generates corresponding feature weights and keyword sets according to the importance of different fields.

[0067] When the analysis module 140 receives a detection command and needs to analyze the objects carried in the hand image set, it first performs preprocessing such as binarization, noise reduction, and tilt correction on the image area of ​​the object. Then, it calls a text extraction algorithm (such as an OCR engine) to recognize the visible printed text, label characters, etc. on the surface of the object and generates text data.

[0068] Subsequently, based on the preset matching rules, the preset battery text feature data corresponding to the text data is retrieved from the battery text feature database. Combining factors such as keyword hit rate, field matching degree, and string similarity, the correlation between the text data and the battery text features is calculated, and the text judgment probability value is output in numerical form.

[0069] The probability value of the text judgment is compared with a preset first threshold. When the probability value of the text judgment is greater than the first threshold, it is determined that the object is highly consistent with the characteristics of a two-wheeled vehicle battery at the text feature level. A battery confirmation result is generated, and the battery confirmation result and the corresponding confidence level are written into the analysis data as a direct basis for whether to issue a warning command in the future.

[0070] When the probability value of text identification is less than the first threshold, the system considers the current text features insufficient to confirm that the object is a battery. It then generates text matching failure data and writes the text matching failure mark into the analysis data to prompt the subsequent process to combine other methods such as shape pattern matching and magnetic field detection to further identify the object's identity, thereby completing the multi-source feature analysis of the object.

[0071] In the hand-carried object recognition link, a battery text feature database and the calculation of text judgment probability values ​​are introduced. Taking advantage of the objective feature that electric vehicle battery shells are generally marked with text information such as "voltage, power, manufacturer information, and safety warnings", the text on the surface of the carried object is extracted from ordinary image texture and processed in a structured manner. By comparing it with the preset battery text feature data in the database, more accurate recognition conclusions can be obtained based solely on the surface text and parameter content when the shape of the carried object is similar to that of a suitcase, toolbox, etc., effectively reducing the risk of misjudgment caused by relying solely on appearance judgment.

[0072] Meanwhile, by setting a first threshold and distinguishing between "confirmed battery results" and "text comparison failure data", the text recognition results have a clear level of credibility in the system. When the text judgment probability value is insufficient, the system does not rashly conclude that it is a battery, but marks the result as a failure and submits it to the subsequent pattern matching or magnetic field detection module for further verification.

[0073] It achieves hierarchical and fault-tolerant identification decisions, which helps maintain the robustness and reliability of the overall system identification under adverse conditions such as complex lighting, partial occlusion, and tag wear. It improves the accuracy and completeness of detecting the behavior of disassembling electric vehicle batteries and further reduces the probability of missing detection of battery thermal runaway risk in elevators.

[0074] In at least one embodiment of this application, the specific steps of executing the detection instruction, parsing the objects in the hand image set, and generating analysis data further include: S207. Analyze the surface pattern on the object carried in the hand image set to generate shape pattern data; S208. Based on a preset battery image library, the shape pattern data is retrieved to obtain a set of suspected battery images. S209. The shape pattern data is compared sequentially with each image in the suspected battery image set; S210. If the comparison result exceeds the second threshold, a confirmed battery result is generated, and analysis data is generated.

[0075] In at least one embodiment of this application, the step of sequentially comparing the shape pattern data with each image in the suspected battery image set, and generating a confirmed battery result and analysis data if the comparison result exceeds a second threshold, further includes: S211. If the comparison result does not exceed the second threshold, then image comparison failure data is generated, and analysis data is generated based on the image comparison failure data.

[0076] Please refer to Figure 3-4 In this embodiment, after the judgment module 130 has determined that the passenger's hands are carrying items and outputs the image area of ​​the carried items, the analysis module 140 first performs preprocessing on the image area of ​​the carried items, such as grayscale conversion, edge enhancement, and distortion correction. Features such as the outline shape of the carried item's shell, the outline of the surface sticker, typical icons, and brand logo patterns are extracted to form shape pattern data that characterizes the appearance features of the carried items.

[0077] Subsequently, the analysis module 140 calls a pre-built battery image library, which stores reference images and feature vectors of batteries for various models of two-wheeled vehicles. Based on the shape pattern data, feature retrieval is performed in the battery image library to select several battery reference images with high similarity to the current shape pattern data, forming a set of suspected battery images after retrieval.

[0078] Subsequently, the analysis module 140 performs a detailed comparison between the shape pattern data and each image in the suspected battery image set. For example, it calculates comprehensive similarity indicators such as contour overlap, key point matching number, and feature vector similarity. When the comparison result exceeds the preset second threshold, it is determined that the object is sufficient to be confirmed as belonging to the electric vehicle battery category in terms of appearance pattern. A confirmed battery result is generated, and the confirmed result and the corresponding similarity value are written into the analysis data.

[0079] When the comparison results between the shape pattern data and any reference image in the suspected battery image set do not exceed the second threshold, the analysis module 140 generates image comparison failure data, writes the image comparison failure mark into the analysis data, and indicates that the current shape pattern features are insufficient to support the battery confirmation conclusion, so as to make a comprehensive judgment in combination with other information such as text recognition results and magnetic field detection results.

[0080] By extracting the surface patterns and shape features of hand-held objects, the appearance information of the objects is converted into comparable shape pattern data. This data is then matched with multiple battery reference appearances in a pre-set battery image library in two levels. The image library is used to first filter out suspected battery image sets, narrowing the search range for subsequent fine comparisons, reducing computational load, and adapting to the real-time processing needs of elevator controllers or edge devices.

[0081] By setting a second threshold, highly similar-looking items can be distinguished from ordinary suitcases, toolboxes, and other objects with similar appearances. Even in cases of illegally modified batteries, partially torn labels, or illegible text, a relatively reliable judgment can still be made based on appearance features such as shell outline, sticker layout, and brand patterns. This compensates for the deficiency that text recognition alone cannot cover batteries without significant text markings.

[0082] In at least one embodiment of this application, the step of generating a warning instruction if the item carried in the analyzed data is a battery further includes: S301. If the condition for confirming that the carried item is a battery is not met, a supplementary monitoring command is generated. S302. Execute the supplementary monitoring command to obtain magnetic field change data; S303. Based on the magnetic field change data, determine whether the object being carried is a battery. If it is a battery, generate a warning command.

[0083] In at least one embodiment of this application, the specific steps of executing the supplementary monitoring command to obtain magnetic field change data of items carried by the passenger include: S304. Acquire data on the change in magnetic field emitted by the excitation coil detected by the detection coil near the elevator door during the process of a passenger passing through the elevator door. S305. Obtain the shape image of the object carried in the hand image set, and map the magnetic field change data onto the shape image to obtain the magnetic induction image of the object; S306. Analyze the magnetic induction range in the magnetic induction image of the object and generate a magnetic induction percentage value; S307. Compare the magnetic induction ratio with the third threshold. If it exceeds the third threshold, generate a warning command.

[0084] Please refer to Figure 5 In this embodiment, when the analysis data formed based on text features, shape and pattern features, etc., fails to meet the preset confirmation conditions, the system determines that the current visual recognition result does not have sufficient confidence, and the processing unit generates a supplementary monitoring command and sends it to the magnetic field detection unit.

[0085] The magnetic field detection unit includes an excitation coil and a detection coil located near the elevator door. The excitation coil outputs an alternating current at a preset frequency to form an alternating magnetic field in the area near the elevator door. The detection coil is used to collect the magnetic field response signal in this area.

[0086] When a passenger carrying an item passes through the elevator door, if the item contains a large amount of metal conductor (such as the casing of a two-wheeled vehicle battery pack and its internal busbars), it will generate a significant eddy current effect on the alternating magnetic field, thereby causing changes in the amplitude, phase, or equivalent impedance of the output signal of the detection coil.

[0087] Triggered by a supplementary monitoring command, the processing unit collects the output of the detection coil as the passenger passes through the elevator door, obtaining corresponding magnetic field change data. Simultaneously, the processing unit extracts the shape image of the object being carried from the hand image set. Based on the outline position and area of ​​the object in the image from the elevator doorway perspective, it constructs a two-dimensional shape mask of the object. Then, it establishes a spatial mapping relationship between the collected magnetic field change data and the object shape image, obtaining a magnetic induction image characterizing the strength of the magnetic response of the object at various locations.

[0088] Subsequently, the processing unit identifies regions with significant magnetic response in the magnetic induction image, calculates the area ratio of these regions within the outline region of the object, and obtains the magnetic induction ratio value.

[0089] The magnetic induction ratio is compared with a preset third threshold. When the magnetic induction ratio exceeds the third threshold, it indicates that a large proportion of the object within the overall outline responds significantly to the alternating magnetic field, meeting the typical characteristics of metal quantity and distribution inside an electric vehicle battery. At this time, the system outputs the determination result that the object is a battery and generates a warning command, driving the elevator control system to trigger voice or display prompts, and restricting elevator operation if necessary.

[0090] If the magnetic induction ratio does not exceed the third threshold, a mark indicating that the magnetic field detection is not confirmed can be recorded in the analysis data.

[0091] Based on the magnetic field detection path of the excitation coil and the detection coil, and taking advantage of the physical characteristics of the large number and volume of metal conductors inside the two-wheeled vehicle battery, the system can still determine whether the carried item has a large metal structure like a battery by measuring the strength and coverage of the magnetic field response, even when the camera's field of view is partially obstructed, the carried item is wrapped in a cloth bag or cardboard box, or the text or patterns on the surface are unclear. This makes up for the missed detection problem caused by obstruction, lighting, and changes in viewing angle in pure visual recognition.

[0092] By combining magnetic field change data with images of the object's shape, a magnetic induction image is constructed and the magnetic induction ratio is calculated. This makes the judgment based not only on whether there is metal in the detection area, but also on the degree of coverage of the metal response within the outline of the object. This effectively distinguishes between small metal items (such as keys, mobile phones, and metal tools) and battery packs with a large amount of metal conductors inside, reducing the false alarm rate.

[0093] By hierarchically fusing image recognition results with magnetic field detection results, supplementary monitoring commands are only activated when features such as text and patterns cannot be confirmed. This further reduces the risk of batteries being removed and brought into elevators without being identified, significantly improving the reliability and completeness of safety monitoring of battery-carrying behavior in elevator scenarios.

[0094] It should be noted that the first threshold is used to characterize the confidence threshold of the text recognition result. When the text judgment probability value reaches or exceeds the first threshold, it is considered that the object meets the battery feature requirements at the text feature level.

[0095] The second threshold is used to characterize the appearance similarity between the shape pattern of the object and reference images in the battery image library. When the comparison result between the shape pattern data and the suspected battery image reaches or exceeds the second threshold, the object is considered to meet the battery feature requirements at the appearance pattern level.

[0096] The third threshold is used to characterize the coverage of the metal response inside the carrier. When the magnetic induction ratio reaches or exceeds the third threshold, the carrier is considered to meet the metal volume characteristics of the battery pack in terms of magnetic field response characteristics.

[0097] The first threshold, the second threshold, and the third threshold are manually set thresholds.

[0098] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A safety monitoring system for battery-carrying behavior in two-wheeled vehicles based on posture recognition, characterized in that, The posture recognition-based two-wheeled vehicle battery carrying behavior safety monitoring system includes: The image acquisition module is used to acquire video image sequences during the process of passengers entering the elevator; The recognition module is used to identify the posture data of passengers in a video image sequence; The judgment module determines whether the hands of the passenger in the video image sequence are carrying items based on the posture data. If items are carried, the image area corresponding to the items is used as the detection object and a detection instruction is generated. The analysis module executes the detection command to perform image analysis on the object in the video image sequence and generate analysis data. If the object in the analysis data is a battery, a warning command is generated.

2. The two-wheeled vehicle battery-carrying behavior safety monitoring system based on posture recognition according to claim 1, characterized in that, The analysis module includes: The text extraction module is used to extract text data from the surface of the object and parse the text data to obtain the parsed text data. The shape pattern extraction module is used to extract the shape pattern data of the surface of the object being carried.

3. The two-wheeled vehicle battery-carrying behavior safety monitoring system based on posture recognition according to claim 2, characterized in that, The analysis module also includes: The matching module performs battery pattern matching on the shape pattern data. If a match is found, pattern matching data is generated. The results output module generates analysis data based on the parsed text data and / or the pattern matching data.

4. A method for safety monitoring of battery-carrying behavior in two-wheeled vehicles based on posture recognition, applied to the safety monitoring system for battery-carrying behavior in two-wheeled vehicles based on posture recognition as described in any one of claims 1 to 3, characterized in that, The safety monitoring method for battery-carrying behavior in two-wheeled vehicles based on posture recognition includes: Acquire video image sequences of passengers entering the elevator; The video image sequence is parsed to generate a set of images of the passenger's hands, and the set of hand images is used to determine whether the passenger is carrying any items. If the passenger is carrying any items, a detection command is generated. The detection command is executed to analyze the objects carried in the hand image set and generate analysis data. If the item carried in the analyzed data is a battery, a warning instruction is generated.

5. The method for safety monitoring of battery-carrying behavior in two-wheeled vehicles based on posture recognition according to claim 4, characterized in that, The specific steps of executing the detection instruction, parsing the objects carried in the hand image set, and generating analysis data include: Establish a database of battery textual features; The text on the object being carried in the hand image set is analyzed to generate text data; Determine the correlation between the text data and the preset battery text feature data in the battery text feature database, and generate a text judgment probability value; The text judgment probability value is compared with a first threshold. If the text judgment probability value is greater than the first threshold, a battery confirmation result is generated, and analysis data is generated.

6. The method for safety monitoring of battery-carrying behavior in two-wheeled vehicles based on posture recognition according to claim 5, characterized in that, The specific steps of comparing the text judgment probability value with a first threshold, and generating a confirmed battery result and analysis data if the text judgment probability value is greater than the first threshold, further include: If the probability value of the text comparison is less than the first threshold, text comparison failure data is generated, and analysis data is generated based on the text comparison failure data.

7. The method for safety monitoring of battery-carrying behavior in two-wheeled vehicles based on posture recognition according to claim 4, characterized in that, The specific steps of executing the detection instruction, parsing the objects carried in the hand image set, and generating analysis data further include: The surface pattern on the object carried in the hand image set is analyzed to generate shape pattern data; The shape pattern data is retrieved based on a preset battery image library to obtain a set of suspected battery images. The shape pattern data is compared sequentially with each image in the suspected battery image set. If the comparison result exceeds the second threshold, a confirmed battery result is generated, and analysis data is generated.

8. The method for safety monitoring of battery-carrying behavior in two-wheeled vehicles based on posture recognition according to claim 7, characterized in that, The step of comparing the shape pattern data sequentially with each image in the suspected battery image set, and generating a confirmed battery result and analysis data if the comparison result exceeds a second threshold, further includes: If the comparison result does not exceed the second threshold, image comparison failure data is generated, and analysis data is generated based on the image comparison failure data.

9. The method for safety monitoring of battery-carrying behavior in two-wheeled vehicles based on posture recognition according to claim 4, characterized in that, If the item carried in the analyzed data is a battery, the step of generating a warning instruction further includes: If the condition for confirming that the carried item is a battery is not met, a supplementary monitoring command is generated; Execute the supplementary monitoring command to obtain magnetic field change data; Based on the magnetic field change data, it is determined whether the object being carried is a battery. If it is a battery, a warning command is generated.

10. The method for safety monitoring of battery-carrying behavior in two-wheeled vehicles based on posture recognition according to claim 9, characterized in that, The specific steps for executing the supplementary monitoring command to obtain data on changes in the magnetic field of items carried by the passenger include: Acquire data on the changes in the magnetic field emitted by the excitation coil detected by the detection coil near the elevator door during the process of a passenger passing through the elevator door; Obtain the shape image of the object carried in the hand image set, and map the magnetic field change data onto the shape image to obtain the magnetic induction image of the object; Analyze the magnetic induction range in the magnetic induction image of the object and generate a magnetic induction percentage value; The magnetic induction ratio is compared with a third threshold. If it exceeds the third threshold, a warning instruction is generated.

Images