Vehicle underbody living body detection method, computer program product, electronic device and vehicle

Abstract

The application relates to a vehicle underbody living body detection method, a computer program product, an electronic device and a vehicle. The detection method comprises the following steps: taking a vehicle underbody image as a reference image when the vehicle is powered off; determining a real-time image with a difference region as a to-be-detected image based on the reference image and the real-time image of the vehicle underbody; determining feature detection information of the to-be-detected image, and realizing identification of a vehicle underbody living body and / or determination of a vehicle underbody living body state based on the feature detection information. Based on this, the application can improve the pertinence and efficiency of vehicle underbody detection, and guarantee the accuracy and effectiveness of vehicle underbody living body detection results.

Description

Technical Field

[0001] This application relates to the field of vehicle monitoring technology, and in particular to a method for detecting live objects under a vehicle, a computer program product, electronic equipment, and a vehicle. Background Technology

[0002] The underside of a vehicle is a blind spot, easily becoming a habitat for children and small animals. This poses a double safety hazard, as animals may chew on and damage vehicle parts. The core challenge of existing under-vehicle live animal detection technology is the lack of all-weather, highly reliable live animal detection, and all solutions have significant limitations, making it difficult to meet actual vehicle usage needs.

[0003] In existing technologies, at the perception level, they are susceptible to interference from low light and complex backgrounds; thermal imaging sensors are greatly affected by ambient temperature; optical cameras have poor signal-to-noise ratios in low light; and ultrasonic sensors are prone to measurement deviations. At the analysis level, they lack the ability to distinguish between targets that pass by briefly and those that linger for a long time, and their decision-making logic is simplistic. At the execution level, the response mechanism is not precise enough, and some solutions are prone to false triggering of deflection actions. In addition, some solutions rely on network communication, and their reliability decreases in environments with poor signal, such as underground parking lots. The technical specifications of the perception modules in some solutions are unclear, and they cannot distinguish between biological species, resulting in insufficient overall detection accuracy and stability. Summary of the Invention

[0004] This application provides a method for detecting live bodies under a vehicle, a computer program product, an electronic device, and a vehicle, which can solve at least one technical problem in the prior art.

[0005] Accordingly, this application provides a method for detecting liveness under a vehicle. The method includes: when the vehicle is powered off, using an image of the undercarriage as a reference image; based on the reference image and a real-time image of the undercarriage, determining the real-time image with a difference region as an image to be detected; determining feature detection information of the image to be detected, and realizing the identification of liveness under the vehicle and / or the determination of the liveness status of the undercarriage based on the feature detection information.

[0006] In one embodiment of this application, the feature detection information includes at least the category, location information, and confidence information of the target under the vehicle. The step of identifying a live object under the vehicle and / or determining the state of the live object based on the feature detection information includes: identifying the live object under the vehicle based on the category of the target, the location information, and the confidence information; and determining the state of the live object under the vehicle based on the change in the location information when it is determined that a live object exists under the vehicle, wherein the state includes entering or leaving the vehicle.

[0007] In one embodiment of this application, the detection method further includes: marking the living body under the vehicle when it is determined that there is one under the vehicle; or recording the time when the living body leaves the vehicle and / or enters the vehicle; or triggering a preset early warning response strategy to encrypt and store the information related to the early warning event.

[0008] In one embodiment of this application, the early warning response strategy is determined based on the category and dwell time of the living object under the vehicle; when the living object under the vehicle is a first preset category and the dwell time is less than a preset threshold, the vehicle's audio-visual warning is triggered; when the living object under the vehicle is a first preset category and the dwell time exceeds the preset threshold, or is a second preset category, the vehicle's local audio-visual warning is triggered and a pop-up warning is sent to the terminal bound to the vehicle; when the living object under the vehicle is a third preset category, while sending a pop-up warning to the terminal bound to the vehicle, the vehicle's start is restricted until the living object under the vehicle leaves.

[0009] In one embodiment of this application, the warning response strategy is adjusted based on the current state of the vehicle; when the vehicle is in a standby state, the vehicle is restricted from starting; when the vehicle is in a driving state, the response level is reduced and a pop-up warning is sent to the terminal bound to the vehicle; when the vehicle is locked and stationary, the warning response strategy is executed based on the preset category of the living body under the vehicle and the response rules corresponding to the dwell time.

[0010] In one embodiment of this application, determining the real-time image with a difference region as the image to be detected based on the reference image and the real-time image of the vehicle underside includes: determining the lighting environment of the real-time image based on image brightness feature parameters; if the real-time image is in a dark lighting environment, performing image enhancement processing on the real-time image; and determining the real-time image with a difference region as the image to be detected based on the image-enhanced real-time image, the unprocessed real-time image in a bright lighting environment, and the reference image.

[0011] In one embodiment of this application, determining the real-time image with a difference region as the image to be detected based on the reference image and the real-time image of the vehicle underside includes: subtracting the reference image and the real-time image pixel by pixel and taking the absolute value to generate an absolute difference map; converting the absolute difference map into a binary image by using a preset grayscale threshold; and processing the binary image to obtain effective connected regions, thereby determining the real-time image containing the effective connected regions as the image to be detected.

[0012] In one embodiment of this application, the reference image and the real-time image of the vehicle underside are used to determine the real-time image with a difference region as the image to be detected. This includes preprocessing the reference image and the real-time image respectively. The preprocessing includes at least grayscale conversion, Gaussian filtering for noise reduction, and size normalization.

[0013] Accordingly, this application provides a computer program product that stores instructions that, when executed by a computer, cause the computer to perform the aforementioned liveness detection method.

[0014] Accordingly, this application provides an electronic device, including: a memory storing a computer program thereon; and a processor for executing the computer program in the memory to implement the above-described liveness detection method.

[0015] Accordingly, embodiments of this application provide a vehicle, including: the electronic device described above; or, a processor, the processor being used to execute the liveness detection method described above.

[0016] The vehicle undercarriage liveness detection method, computer program product, electronic device, and vehicle provided in this application acquire an image of the vehicle undercarriage when the vehicle is powered off as a reference image. The reference image is compared with a real-time image of the vehicle undercarriage to determine the image to be detected containing the difference region. Feature detection information of the image to be detected is extracted and used to identify liveness undercarriage and / or determine the state of liveness undercarriage. This can provide accurate image references that fit the actual use scenario of the vehicle for vehicle undercarriage liveness detection, effectively filter out redundant image information without difference, accurately lock the suspected liveness region, improve the targeting and efficiency of vehicle undercarriage detection, and at the same time, rely on feature detection information to achieve accurate identification and / or state determination of liveness undercarriage, ensuring the accuracy and effectiveness of vehicle undercarriage liveness detection results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating one embodiment of the liveness detection method of this application;

[0019] Figure 2 This is a schematic diagram of the camera installation position and field of view in one embodiment of this application;

[0020] Figure 3 This is a schematic diagram of the undercarriage liveness detection system module provided in an embodiment of this application;

[0021] Figure 4 This is a flowchart illustrating one embodiment of step S200 of this application;

[0022] Figure 5This is a flowchart illustrating another embodiment of step S200 of this application;

[0023] Figure 6 This is a flowchart illustrating one embodiment of step S300 of this application;

[0024] Figure 7 This is a schematic diagram of the structure of the electronic device of this application. Detailed Implementation

[0025] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be noted that the following embodiments are merely illustrative of the present application and do not limit its scope. Similarly, the following embodiments are only some, not all, embodiments of the present application, and all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0026] It should be understood that the terms "upper," "lower," "left," "right," "front," "back," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or relative positional relationship shown in the accompanying drawings. They are used solely for the convenience of describing this application and for simplification, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Unless otherwise specified, the above-mentioned orientational descriptions can be flexibly set in practical applications, provided that the relative positional relationships shown in the accompanying drawings are satisfied.

[0027] The terms "first" and "second" are configured for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0028] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "communication" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection. They can refer to a direct connection or an indirect connection through an intermediate medium, or a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0029] In embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, article, or apparatus that includes that element.

[0030] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0031] The following is a detailed analysis of the proposed solution with reference to the accompanying drawings:

[0032] Please see Figure 1 , Figure 1 This is a flowchart illustrating one embodiment of the vehicle underbody liveness detection method of this application, as shown below. Figure 1 The detection method provided in this application includes the following steps:

[0033] S100 uses the image of the underside of the vehicle as a reference image when the vehicle is locked.

[0034] Please combine further Figure 2 and Figure 3 , Figure 2 This is a schematic diagram showing the camera installation position and field of view in one embodiment of this application. Figure 3 This is a schematic diagram of the vehicle underbody liveness detection system module provided in an embodiment of this application.

[0035] like Figure 3 The vehicle undercarriage liveness detection system 100 in this application includes: an image acquisition module 110, an image processing module 120, an improved YOLOv8 target detection module 130, an early warning module 140, and a storage module 150.

[0036] like Figure 2As shown, the image acquisition module 110 includes a front-view camera, a rear-view camera, a left-view camera, and a right-view camera. The field of view of each camera covers the entire circumference of the vehicle body. In this embodiment, the blind spot (a small area of ​​ground around the vehicle body) is considered part of the vehicle's underside. The image acquisition module 110 acquires images (video frames) of the ground around the vehicle body in real time, and uses the video frames of the ground around the vehicle body when the vehicle is powered off and locked as the reference image. In specific implementation, when the user powers off and locks the vehicle, the vehicle underside liveness detection system of this application is activated, and the vehicle cameras begin to acquire video images of the ground around the vehicle body, using the first frame as the reference image.

[0037] Understandably, in this embodiment, using the first frame image at the moment the car is locked as the reference image ensures a clean background free of target objects, preventing subsequent frames from being mixed with initial interference such as fallen leaves or passing animals, which could lead to missed detections in later comparisons. Simultaneously, this fixed reference image avoids the drawback of existing technologies that continuously update the background model, easily misinterpreting slow environmental changes such as gradual changes in day and night lighting or dust accumulation as target signals. Subsequent real-time frames only need to be compared with this initial background to identify new changes, ensuring detection reliability.

[0038] Furthermore, fixing the first frame image as a baseline eliminates the need for continuous model updates, significantly reducing the power consumption of the vehicle system, alleviating computational load, extending battery life, and improving response speed. Moreover, this baseline image corresponds to the starting point of the vehicle's parking cycle, ensuring that all real-time detection comparison standards are consistent and avoiding confusion in detection results caused by an unstable baseline.

[0039] The above implementation uses a clean image at the time of vehicle locking as a fixed benchmark, avoiding the shortcomings of continuously updated background models that are susceptible to interference from day-night cycles and weather changes. This provides a stable and reliable reference for subsequent differential comparison, ensuring the anti-interference capability of the entire detection process. Simultaneously, it reuses the vehicle-mounted surround-view camera, eliminating the cost and modification process of adding dedicated sensors, and includes blind spots in the detection scope, achieving full coverage detection of the area under the vehicle.

[0040] S200 uses a reference image and a real-time image of the underside of the vehicle to identify real-time images of areas with discrepancies as the images to be detected.

[0041] Please combine further Figure 4 , Figure 4 This is a flowchart illustrating an implementation method of step S200 of this application, as shown below. Figure 4 Step S200 further includes the following sub-steps:

[0042] S210: Subtract the reference image from the real-time image pixel by pixel and take the absolute value to generate an absolute difference map.

[0043] This application employs the inter-frame difference method, a target detection algorithm based on computer vision. The principle is to calculate the difference in grayscale values ​​of corresponding pixels between two specified frames, extracting the changed regions in the images, thereby achieving the detection and localization of new targets. In this embodiment, the first fixed reference image acquired after the vehicle is locked is used as the sole reference, and pixel-level difference operations are performed frame-by-frame with each subsequent real-time acquired image.

[0044] The algorithm first preprocesses the reference image and the real-time image of the vehicle's underside. This preprocessing includes at least grayscale conversion, Gaussian filtering for noise reduction, and size normalization to eliminate image noise and size differences. Then, it calculates the grayscale difference between corresponding pixels in the two images and takes the absolute value to generate an absolute difference map that visually reflects the changed area. Next, a threshold is set to convert the difference map into a black-and-white binary image, completely separating the changing foreground area (which may contain targets) from the unchanging background area. Finally, a closing operation is used to repair holes and breaks in the foreground area, filtering out the difference areas corresponding to valid contours. This algorithm has simple computational logic and high efficiency. In this scheme, it can quickly filter out unchanged background images and accurately locate new targets under the vehicle and on the surrounding ground. Furthermore, the design combined with a fixed reference image avoids the environmental interference problems caused by continuous background model updates in traditional background subtraction methods.

[0045] Grayscale conversion refers to the process of converting a color image into a grayscale image, reducing computational load and highlighting brightness differences. Furthermore, for color video frames captured by vehicle-mounted cameras (both the baseline and real-time images are in color), each pixel contains brightness values ​​(0-255) for three channels: red (R), green (G), and blue (B). Grayscale conversion can employ a weighted average method to fuse the three channels into a single grayscale value, ensuring that each pixel retains only a brightness information value of 0-255, thus converting the color image into a grayscale image. This operation compresses the image data to one-third of its original size, significantly reducing the computational load of subsequent pixel-level operations, while also eliminating the interference of color information, focusing on the brightness differences between the ground under the vehicle and the living target, making the core features of subsequent differential comparison more prominent.

[0046] Furthermore, denoising refers to the operation of smoothing images using algorithms such as Gaussian filtering to eliminate minute noise interference. Specifically, when reference images and real-time images are acquired in a vehicle environment, they are easily affected by electromagnetic interference from vehicle circuits, minor fluctuations in ambient light, and noise inherent in the camera hardware itself, resulting in meaningless minute noise such as salt-and-pepper noise and Gaussian noise. In this embodiment, a Gaussian filtering algorithm is used to perform denoising. A 3×3 or 5×5 filtering window is set with each pixel as the center. Different weights are assigned to the pixels within the window using a Gaussian kernel function, and a weighted average is calculated and used to replace the original pixel value, thus achieving image smoothing. This operation can effectively eliminate minute noise in the image, avoid such meaningless pixel changes being misjudged as valid difference regions in subsequent difference operations, and at the same time preserve the contour integrity of the live target, ensuring the accuracy of the difference comparison.

[0047] Furthermore, the size normalization operation ensures that the two images are the same size, facilitating pixel-level calculations. In a specific implementation, due to factors such as slight parameter adjustments after the vehicle camera locks and mechanical jitter during acquisition, the reference image and the real-time image may have slight deviations in pixel resolution and overall image size. This application can use a bilinear interpolation algorithm to uniformly process the reference image and the real-time image, scaling both frames to a preset fixed pixel size. If there are image edge deviations, adaptive cropping or edge patching is performed simultaneously. In this way, it can be ensured that the number of pixel rows and columns of the reference image and the real-time image are completely consistent, allowing the pixels of the two images to correspond one-to-one, providing a basis for subsequent pixel-by-pixel subtraction difference calculations, and avoiding comparison failures caused by size deviations.

[0048] The above implementation method, through a standardized preprocessing procedure, eliminates interference factors such as image noise and size differences, improving the accuracy of pixel-level differential comparison. Furthermore, binarization and closing operation repair effectively solve the problem of missed detections caused by broken target contours. Simultaneously, only image frames containing valid differences are retained, filtering out a large amount of invariant and invalid data, significantly reducing the computational load on subsequent target detection modules and improving the real-time response speed of the entire system.

[0049] Furthermore, pixel-by-pixel operations are performed on the grayscale images of the reference image and the real-time image after grayscale conversion, denoising, and size normalization. The pixel grayscale value at the same coordinate position in the two images is used as the calculation object. The pixel grayscale value at that position in the real-time image is subtracted from the pixel grayscale value at the corresponding position in the reference image, and the absolute value is taken to obtain the difference pixel value at that coordinate position.

[0050] Furthermore, after sequentially traversing all pixels of both images and performing the aforementioned operations, a single-channel absolute difference image is generated. The pixel values ​​of the grayscale image range from 0 to 255, therefore the pixel values ​​of the absolute difference image also fall within this range. A pixel value of 0 indicates that the pixel at that location shows no change between the reference and real-time images, while a larger pixel value indicates a more significant grayscale change at that location. Visually, brighter areas in the absolute difference image represent a greater degree of pixel change compared to the reference image. Areas with live objects or foreign targets such as those under the vehicle and on the surrounding ground will appear bright due to changes in grayscale characteristics, while the unchanged background areas will appear almost pure black. This achieves preliminary quantization and visualization of the target change areas and background areas, providing a clear computational foundation for subsequent binarization processing.

[0051] S220 converts the absolute difference image into a binary image by using a preset grayscale threshold.

[0052] Furthermore, the absolute difference image obtained through pixel-by-pixel calculation is converted into a black-and-white binary image, and the absolute difference image is converted into a binary image based on a preset grayscale threshold.

[0053] The preset grayscale threshold can be calibrated based on the actual scene of vehicle undercarriage detection, adapting to the grayscale variation characteristics of the ground and living objects. Then, all pixels in the absolute difference image are processed according to this preset grayscale threshold. Pixels with values ​​higher than the preset grayscale threshold are uniformly set to white (e.g., identified as foreground areas, pixel value 255), representing areas with significant grayscale changes compared to the reference image. Pixels with values ​​lower than the preset grayscale threshold are uniformly set to black (e.g., identified as background areas, pixel value 0), representing areas with no significant grayscale changes compared to the reference image.

[0054] In the above implementation, by discarding the blurred intermediate grayscale change information in the absolute difference image and retaining only the significant grayscale change features, a clear and thorough separation of the foreground change area and the background unchanged area is achieved. This not only eliminates the weak and invalid changes caused by minor fluctuations in illumination and slight reflections on the ground, but also accurately locates the significant change areas suspected of containing living objects / foreign objects. Furthermore, it simplifies the image into a binary form containing only two pixel values, 0 and 255, greatly reducing the computational complexity of subsequent closing operations and contour filtering, and laying a simple and clear image foundation for the subsequent identification and extraction of effective contours.

[0055] S230, the binary image is processed to obtain the effective connected regions, so as to determine the real-time image containing the effective connected regions as the image to be detected.

[0056] Furthermore, a closing operation is performed on the thresholded black-and-white binary image. This operation uses a morphological operation logic of dilation followed by erosion to fill and connect sparse breaks, small cracks, and internal voids in the foreground white region of the binary image. This restores the discrete and incomplete foreground region into a full, coherent, and fully connected region, effectively compensating for the foreground region breakage caused by uneven lighting, blurred target edges, and other factors. This avoids missing or discontinuous target contours during subsequent contour extraction, ensuring the integrity of the target contour.

[0057] Furthermore, after completing the closing operation, contour detection and filtering are performed on the processed binary image. The contour extraction algorithm identifies all closed foreground connected region contours in the image, and based on the actual scene of vehicle undercarriage detection, effective contour judgment criteria are set, such as thresholds for contour area, perimeter, aspect ratio, etc., to determine whether there are valid contours in the image that meet the criteria.

[0058] If no valid contours are found after detection, it is determined that there is no actual target change in the real-time image, and it is marked as an indifferent frame and deleted directly, and will not enter the subsequent detection process.

[0059] If a valid contour is detected, the real-time image is determined to contain a suspected live / foreign object target. The original real-time image of the corresponding scene is then selected as the image to be detected and pushed to the subsequent target detection module for further accurate identification.

[0060] In the above embodiments, by performing a series of operations such as subtracting absolute values ​​pixel by pixel to generate an absolute difference map, binarizing with a preset grayscale threshold, and extracting effective connected regions, the significant change regions between the reference image and the real-time image under the vehicle can be accurately separated. This effectively filters out redundant pixel interference with no difference between the images, improves the accuracy of locating the difference regions under the vehicle and the efficiency of determining the image to be detected, and provides a reliable image basis for the accurate identification of living objects under the vehicle.

[0061] Please combine further Figure 5 , Figure 5 This is a flowchart illustrating another embodiment of step S200 of this application, as shown below. Figure 5 Step S200 further includes the following sub-steps:

[0062] S201, determine the lighting environment of the real-time image based on the image brightness feature parameters.

[0063] Understandably, images of the underside of a vehicle in low-light environments are prone to uneven illumination, blurred edges, and loss of detail. Directly extracting features from these images would severely impact the accuracy of subsequent liveness detection. In this embodiment, to improve the accuracy of subsequent liveness detection, it is necessary to first determine the lighting environment of the real-time image based on image brightness feature parameters.

[0064] Specifically, the system first extracts the global brightness feature parameters of the real-time image of the vehicle under the vehicle to be detected. By traversing all pixels of the real-time image, the average gray value of the entire image is calculated. The gray value ranges from 0 to 255, where 0 represents pure black and 255 represents pure white. The average gray value directly reflects the overall brightness level of the image.

[0065] Meanwhile, taking into account the special characteristics of the vehicle under-car detection area, the brightness of key detection areas under the vehicle (such as gaps in the chassis and around the wheels, areas where live objects are likely to hide) is sampled and statistically analyzed in different regions to obtain the local brightness distribution of the image, thus avoiding the deviation in the judgment of a single global brightness index caused by local shadows and reflections under the vehicle.

[0066] Subsequently, the calculated global average grayscale value and regional brightness statistics are compared with a pre-calibrated illumination judgment threshold for the undercarriage detection scenario (this threshold is trained and calibrated using a large number of samples of bright and dark undercarriage scenes and can be flexibly adjusted according to the actual application scenario). If the global average grayscale value and regional brightness statistics of the real-time image are both higher than the preset illumination judgment threshold, the environment in which the real-time image is located is determined to be a bright environment; if the above brightness indicators are lower than the preset illumination judgment threshold, it is determined to be a dark environment. This completes the accurate and efficient determination of the bright environment of the real-time image, providing a clear basis for determining whether to perform image enhancement operations in the future.

[0067] S202, If the real-time image is in a low-light environment, perform image enhancement processing on the real-time image.

[0068] Furthermore, if the real-time image is in a low-light environment, in order to improve the problems of loss of details, blurry image and low contrast in the real-time image of the undercarriage under light, and to avoid missed or misjudged detection of subsequent difference areas due to poor image quality, the system immediately performs targeted image enhancement processing on the real-time image of the undercarriage under the vehicle.

[0069] The image enhancement process in this solution consists of three core steps: illumination component estimation, chromatic aberration restoration, and post-processing.

[0070] The illumination component estimation operation uses Gaussian filters to blur the original image at different scales to simulate the illumination distribution. The filtering results at different scales are then weighted and averaged to obtain an illumination estimate that is both globally smooth and preserves local edges. The specific formula is as follows:

[0071]

[0072] in, The coordinates of the enhanced output image, Indicates the quantity of scale, Indicates the first ( The weights of each scale, Represents the coordinates of the input image. Indicates the first Gaussian wrapping function of scale This indicates a convolution operation.

[0073] The result obtained from the above calculation It is usually grayscale, and the colors are severely distorted.

[0074] Therefore, a color restoration factor is needed to recover the color information of the original image. The color restoration factor is typically based on the ratios of the original image's channels (R, G, B), and the specific formula is as follows:

[0075]

[0076]

[0077] in, Represents the original observed image. Coordinates on each color channel Pixel value at that location, Indicates the same pixel point The sum of pixel values ​​for all color channels. This indicates three color channels. This represents the gain coefficient, which enhances overall color saturation. This represents the intensity control factor, used to scale the logarithmic result and control the intensity of color reproduction. Indicates the use of the first Color recovery factor for each channel, Indicates the first The coordinates of the enhanced output image for each channel. This indicates the final output number. Images with multiple channels.

[0078] Finally, post-processing operations, such as linear or non-linear stretching, map pixel values ​​to the standard display range (0-255) to enhance low-light images.

[0079] Understandably, this application does not require replacing the high-end night vision camera. By simply adjusting the algorithm and hardware parameters, ordinary optical cameras can overcome the limitations of lighting conditions and maintain stable and reliable vehicle under-vehicle detection capabilities in various low-light / dark-light conditions such as nighttime, underground garages, tunnels, and rainy / foggy conditions. This significantly reduces the hardware deployment and modification costs of the system and improves the practicality and versatility of the solution.

[0080] Secondly, image enhancement achieved by automatically adjusting sensor parameters and signal amplification coefficient can accurately improve detail while suppressing image noise in low-light environments. This allows the enhanced image to clearly restore the contour features and texture details of the target under the vehicle, providing a sufficient and accurate image information foundation for subsequent feature extraction, target detection, and liveness detection. It effectively avoids detection omissions and misjudgments caused by blurry images and loss of details in low light, significantly improving the accuracy and reliability of liveness detection under the vehicle.

[0081] Finally, the system can automatically trigger an enhanced mode based on ambient lighting conditions, requiring no manual intervention throughout the process. This intelligent adjustment adapts to lighting conditions, aligning with unattended vehicle-mounted equipment scenarios and enhancing the system's automation and intelligence. Furthermore, it optimizes camera imaging in low-light environments from the ground up, improving the environmental adaptability and all-scenario coverage of the under-vehicle liveness detection system. This ensures the system can continuously and stably complete under-vehicle detection tasks in both bright daylight and various low-light conditions, guaranteeing all-weather operation and further strengthening the practicality and robustness of the entire under-vehicle liveness detection solution.

[0082] S203, based on the real-time image after image enhancement processing, the real-time image under bright light conditions without processing, and the reference image, the real-time image with the difference region is determined as the image to be detected.

[0083] Furthermore, after completing the enhancement processing of the low-light image, the system uses the unprocessed real-time image and reference image under bright light, as well as the enhanced real-time image of the vehicle's underside, as processing objects. It then combines these with the pre-stored reference image of the vehicle's underside to perform difference region detection, thereby determining the image to be detected.

[0084] S300, determine the feature detection information of the image to be detected, and realize the identification of the living body under the vehicle and / or the determination of the living body status under the vehicle based on the feature detection information.

[0085] In this embodiment, the feature detection information includes at least the category, location information, and confidence level of the target under the vehicle. Furthermore, the unprocessed real-time image and the reference image under bright light are input into the backbone network CSPDarknet for feature extraction.

[0086] Specifically, the process involves first adjusting the original image to be detected under bright light conditions and the image after illumination component estimation, color difference restoration, and post-processing enhancement under low light conditions to a size and channel format suitable for the input requirements of the CSPDarknet network. Then, these images are input into the CSPDarknet backbone network for hierarchical feature extraction. The shallow convolutional layers use small kernels and low strides to focus on pixel-level micro-features, extracting key visual features such as edge contours, surface textures, and color blocks for vehicle undercarriage detection. This provides crucial information for distinguishing live targets from non-live objects like stones and scraps of paper. The deep convolutional layers, after shallow convolution and multiple pooling operations, have a larger receptive field, integrating the scattered micro-details extracted from the shallow layers into semantically meaningful feature representations. This allows for the identification of various semantic features in the vehicle undercarriage scene, such as cat ears, mouse body parts, and wheels. Information is used to upgrade from pixel features to high-level semantic features, providing a key basis for subsequent target classification and localization. The core CSP cross-stage partial connection structure of CSPDarknet, by splitting the feature stream, processing it separately and then fusing it, significantly reduces the computational load under the limited computing power of the vehicle, enriches gradient combinations, effectively avoids the gradient vanishing problem during deep network training, and ensures that the microscopic details of the living object extracted in the shallow layer are not lost in the deep layer operation. It significantly improves the robustness of the network to complex environments such as uneven lighting under the vehicle, diverse target shapes, and cluttered backgrounds. Finally, CSPDarknet outputs a multi-scale feature map that integrates microscopic details of the living object and high-level semantic information, providing a comprehensive and high-quality feature foundation for the subsequent PAN-FPN feature fusion stage.

[0087] Furthermore, PAN-FPN efficiently mixes features at different scales extracted by CSPDarknet. The PAN-FPN structure receives multi-scale feature maps from CSPDarknet and achieves efficient mixing of features at different scales through a bidirectional feature transfer and fusion mechanism. This accurately adapts to the detection needs of live targets under vehicles, which vary in size and shape. The FPN (Feature Pyramid Network) transfers high-level semantic information extracted from deep networks to lower-level feature maps from top to bottom, compensating for the lack of semantic information in lower-level features. This allows lower-level features to accurately distinguish live targets from background interference such as vehicle undercarriage lines and ground textures, greatly improving the ability to identify small, featureless live targets (such as mice).

[0088] The PAN (Path Aggregation Network) then transmits microscopic details (such as liveness edges and textures) from the lower-level features to the higher-level feature maps from the bottom up. This supplements the missing details in the higher-level features, helping them to more accurately pinpoint the outline of the live target, thereby significantly improving the localization accuracy of the bounding box and avoiding issues such as selection offset and missed boxes. Through this bidirectional complementary feature fusion, each scale of feature map ultimately possesses both the high-level semantic information required for accurate recognition and the microscopic details required for accurate localization. This provides comprehensive and high-quality feature support for subsequent classification and regression tasks performed by the decoupled head, ensuring the accuracy and reliability of liveness detection under vehicles.

[0089] Furthermore, the decoupled head performs precise analysis based on the multi-scale feature map fused with PAN-FPN. The core principle is to separate and independently process the two main tasks of target classification and bounding box regression, avoiding the conflict and interference between the two target tasks during single-head joint optimization, thus improving the classification accuracy and localization precision of vehicle underbody liveness detection. Specifically, the classification branch analyzes the region features for each potential target region on the feature map through convolution operations and activation functions, outputting a probability vector for the corresponding category. Each dimension of the vector corresponds to a preset category (such as cat, mouse, non-liveness, etc.), and the dimension with the highest probability value is the predicted category for that region.

[0090] The regression branch focuses on accurate bounding box prediction, outputting the precise center coordinates, width, and height of the target bounding box through fitting operations. Simultaneously, it outputs the object confidence score, which quantitatively represents the probability of a real object existing within the corresponding bounding box, used to filter out false boxes and low-confidence predictions. Finally, the decoupling head integrates the outputs of both branches to generate detection results containing the object's specific category, bounding box coordinates, and corresponding confidence score, providing accurate and reliable criteria for subsequent live target screening and differentiation.

[0091] Please combine further Figure 6 , Figure 6 This is a flowchart illustrating an implementation method of step S300 of this application, as shown below. Figure 6 Step S300 further includes the following sub-steps:

[0092] S310, based on the category, location information and confidence information of the target to be detected, the identification of a living body under the vehicle is completed.

[0093] Specifically, the process begins with an initial screening based on confidence levels to filter out invalid detection results. First, it's determined whether the confidence level of the target to be detected exceeds a preset confidence threshold, for example, 80%. If the confidence level is below the threshold, it's considered an invalid detection, such as noise under a vehicle or false targets caused by light and shadow interference. The corresponding video frame is then deleted, terminating further processing for that target. If the confidence level is above the threshold, it's considered a valid detection target, and the process proceeds to the next step: determining the location range.

[0094] Furthermore, the location range determination identifies targets intruding under the vehicle. Based on the target's location information, such as bounding box coordinates, it is determined whether the target falls within a preset effective monitoring range under the vehicle. This effective monitoring range is a pre-defined image area under the vehicle, such as the image coordinate range corresponding to the vehicle chassis projection. If the core area of ​​the target's bounding box (e.g., center coordinates) or more than 50% of the box area falls within the monitoring range, the object is determined to have entered the vehicle's underside range. Of course, other proportions can be used in other implementations; no specific limitation is made here. If the target does not fall within the vehicle's underside range, it is determined to be an irrelevant target around the vehicle's underside, and the corresponding video frame is directly deleted. If the target does fall within the vehicle's underside range, the core liveness detection process begins.

[0095] Furthermore, the system accurately classifies objects, distinguishing between live and non-live targets. Based on the target's category information, it matches a pre-defined live target category library, improving upon the live target categories already labeled during YOLOv8 training, such as cats, dogs, mice, snakes, and humans, and the non-live target category library, such as stones, paper scraps, plastic bottles, screws, and other common foreign objects under vehicles. If the target category matches the non-live target category library, the corresponding video frame is deleted; if the target category matches the live target category library, the indicated live object under the vehicle is marked. Specifically, this includes:

[0096] First, record the specific category of the live animal, such as cat or dog. Then, assign a unique identification number to the live animal, such as V1, V2, V3 according to the detection sequence, or C1-01 according to the location, forming a basic record of the live animal with category + unique number, in preparation for subsequent status tracking. In this way, the entry / exit status of the same live animal can be identified, avoiding confusion between multiple live animals.

[0097] In the aforementioned real-time method, the under-vehicle liveness detection step relies on the collaborative determination of multiple parameters such as the category, location, and confidence level of the target under the vehicle to accurately distinguish between live and non-live targets from the target in the different areas under the vehicle, effectively eliminating misjudgments caused by interference factors such as debris and shadows, and improving the accuracy and robustness of liveness detection in the under-vehicle scenario.

[0098] S320: When it is determined that there is a live body under the vehicle, the state of the live body under the vehicle is determined based on the change in position information. The state includes entering or leaving the vehicle.

[0099] Furthermore, when it is determined that there is a live body under the vehicle, the state of the live body under the vehicle is determined based on the change in position information. The state includes entering or leaving the vehicle, and the time when the live body leaves the vehicle and / or enters the vehicle is recorded at the same time.

[0100] Specifically, after determining that there is a living body under the vehicle, the system immediately initiates a continuous tracking mechanism for the living body's position information. It collects and records the continuous position coordinates of the living body within the vehicle detection area in real time and in a time sequence, and constructs the movement trajectory of the living body based on the time sequence position information. At the same time, the system pre-defines the boundary range of the effective detection area under the vehicle (covering areas where the living body can hide, such as the chassis, the area around the wheels, and the chassis gaps), compares the position coordinates of the living body at different times with this boundary range in real time, and analyzes the change in position information in combination with the directional characteristics of the movement trajectory.

[0101] Furthermore, if the movement trajectory of the living object shows a change characteristic of gradually moving from outside the effective detection area under the vehicle to within the area, and the subsequent continuous position coordinates are stably within the effective detection area under the vehicle, then the system determines that the state of the living object under the vehicle is that it has entered the vehicle.

[0102] Specifically, if the trajectory of a living object shows a change characteristic of gradually moving from the effective detection area under the vehicle to outside the area, and the final position coordinates completely leave the effective detection area under the vehicle and there is no subsequent position information to return to the area, then the system determines that the state of the living object under the vehicle is that it has left the vehicle.

[0103] In addition, the system will continuously verify the changes in the liveness location information for a short period of time, eliminating false position changes caused by minor errors in image detection or the target briefly passing through the detection area under the vehicle. This avoids misjudging the liveness entry and exit status, ensuring the accuracy and reliability of the status determination results, and providing accurate status basis for subsequent operations such as whether the system triggers warnings or continuous monitoring.

[0104] The above-described implementation method, through continuous tracking of the position information of the living body under the vehicle and analysis of its temporal motion trajectory, combined with precise comparison of the effective detection area boundary under the vehicle, can accurately determine the specific state of the living body entering or leaving the vehicle, effectively solving the technical problem of only detecting the presence of the living body but being unable to determine its dynamic state.

[0105] Meanwhile, by continuously verifying the positional change status over a short period of time, interference from false positional changes caused by minor errors in image detection and the brief movement of the target can be eliminated, avoiding misjudgment of the liveness status and ensuring the accuracy and reliability of the status determination results. In addition, the determination results can provide accurate and effective status basis for the detection system to subsequently trigger warnings and continue monitoring, upgrading the under-vehicle liveness detection from simple existence detection to dynamic status detection, and improving the intelligence level and detection continuity of the entire under-vehicle liveness detection system.

[0106] Furthermore, upon confirming the presence of a living person under the vehicle, a pre-set early warning response strategy is triggered, and the relevant information of the early warning event is encrypted and stored.

[0107] Specifically, in this embodiment, the early warning response strategy is determined based on the category and dwell time of the live object under the vehicle. The system implements a tiered early warning and vehicle start control strategy based on the detection results of the live object under the vehicle. Differential processing is applied according to the preset risk category of the live object and its actual dwell time in the effective detection area under the vehicle. The specific execution rules for each category and duration are as follows:

[0108] The system pre-classifies live objects under the vehicle into three categories based on their risk level and safety hazard severity. It also sets a threshold for determining the duration of a live object's stay under the vehicle. The stay time begins when the live object stably enters the effective detection area under the vehicle. If the live object leaves midway, the timer is reset and counted again, including the following situations:

[0109] When a live object under the vehicle is detected as belonging to the third preset category, the system does not trigger a local audio-visual warning. Instead, it sends a pop-up warning to the vehicle's bound terminal and simultaneously links the vehicle control system to restrict the vehicle from starting. This starting restriction continues until the system detects that the live object under the vehicle has completely left the effective detection area. Furthermore, at this level, a one-click drive-away function is not provided to avoid alerting the target and causing safety risks.

[0110] 1. When a live animal detected under the vehicle falls into the first preset category, such as a common small pet with low safety risk like a cat or dog, and its stay under the vehicle is less than a preset threshold (e.g., 1 hour; other times are acceptable, no specific limitation is made here), the system will only trigger a local audible and visual warning. This will involve a continuous alarm from the vehicle's in-vehicle buzzer and flashing warning lights to alert on-site personnel or the vehicle owner to remove the animal from under the vehicle. The owner can also remotely remove the animal by triggering a one-click removal function through the vehicle's infotainment system or a linked terminal interface, thus avoiding unnecessary and excessive warnings.

[0111] 2. When a live animal under the vehicle is detected as belonging to the first preset category and its dwell time exceeds the preset threshold, it indicates that the live animal has not been removed in time and the safety hazard has increased. Alternatively, if the live animal under the vehicle is detected as belonging to the second preset category, such as stray cats or dogs, small wild animals, or other medium- to high-risk live animals that may cause mechanical failures after the vehicle is started due to their dwell time, the system will immediately send a pop-up warning to the vehicle's bound terminal (owner's mobile APP, vehicle terminal, etc.) while triggering the vehicle's local audible and visual warning. The pop-up will simultaneously display key information such as the category of the live animal under the vehicle, the real-time dwell time, and the specific hiding location under the vehicle. Both the pop-up and the vehicle terminal are equipped with a one-click removal operation entry, allowing the owner to remotely and accurately trigger the removal function. The owner can also remotely be aware of the situation and take appropriate measures in a timely manner.

[0112] 3. When a live object under the vehicle is detected as belonging to the third preset category, such as a person or other living being that poses a significant safety hazard or threat, or is at risk of being deliberately concealed or vandalized, in order to avoid alerting the target and ensure vehicle safety, the system will not trigger a local audible and visual warning on the vehicle. Instead, it will send a pop-up warning to the vehicle's bound terminal containing the live object detection time and the precise location under the vehicle. Simultaneously, the system will immediately activate the vehicle control system and implement software logic restrictions on core starting processes such as engine start and ignition via the vehicle's CAN bus, preventing the vehicle from starting normally. This starting restriction will continue until the system detects in real time that the live object of the third preset category has completely left the effective detection area under the vehicle, and after multiple status verifications confirm that there is no re-entry, the system will automatically lift the vehicle starting restriction and restore the vehicle's normal starting and usage functions.

[0113] Understandably, the aforementioned pre-defined criteria for classifying live animals and the pre-defined threshold for dwell time can be customized by the vehicle owner on the vehicle-binding terminal to adapt to the security needs of different usage scenarios, making the graded early warning and control strategy more flexible.

[0114] In the above implementation, a tiered early warning and vehicle start control strategy is used to classify pre-set categories based on the risk level of live objects under the vehicle. The warning intensity and control measures are dynamically adjusted based on the dwell time. For low-risk, short-duration live objects, only local audio-visual warnings are implemented, effectively avoiding unnecessary interference to users caused by excessive warnings. For medium- to high-risk or low-risk live objects with prolonged dwell time, dual warnings are implemented both locally and at the terminal, ensuring that vehicle owners can be aware of the situation under the vehicle and take appropriate measures regardless of their location. At the same time, for the highest-risk third pre-set category of live objects, strong control measures such as terminal pop-up warnings and vehicle start restrictions are implemented to eliminate major personal safety and vehicle equipment safety hazards from the source. The differentiated design of control measures for each category accurately matches the safety hazard level of different live objects, improving the accuracy of warnings and the targeting of control. Furthermore, the pre-set categories and dwell thresholds can be customized, taking into account the flexibility of the strategy and the actual usage needs of different users. This effectively balances the protection of live objects under the vehicle, the safety of vehicle equipment, and the safety of personnel and property, significantly enhancing the practical application value and all-round safety protection capabilities of the live object detection system under the vehicle.

[0115] Furthermore, in this embodiment, the early warning response strategy is adjusted based on the vehicle's current state. The system dynamically adapts and adjusts the early warning and control operations after under-vehicle liveness detection based on the vehicle's actual operating state. Differentiated execution rules are formulated for three typical vehicle states: waiting to start, driving, and locked and stationary, to ensure the rationality, safety, and practicality of early warning control, as detailed below:

[0116] 1. When the vehicle is in a standby state, restrict the vehicle from starting.

[0117] Specifically, when the system detects a living object under the vehicle and obtains information from the vehicle control system that the vehicle is in a ready-to-start state (such as the owner unlocking the car, pressing the brake, or the vehicle's infotainment system entering the start-ready mode, etc., before ignition), the system immediately activates the vehicle's core control systems, such as power and ignition, to directly restrict the vehicle's starting process, blocking key starting steps such as engine ignition and power output. This avoids damage to the living object under the vehicle from the source and prevents mechanical failures caused by interference from the living object.

[0118] 2. When the vehicle is in motion, reduce the response level and send a pop-up warning to the terminal bound to the vehicle.

[0119] Furthermore, when the system detects a living object under the vehicle while the vehicle is in motion, considering that implementing start restrictions and local audio-visual warnings during driving can easily distract the driver and cause driving safety hazards, the system will proactively reduce the response level and will not perform any local vehicle control or audio-visual warning operations. Instead, it will send a pop-up warning to the terminal bound to the vehicle (owner's mobile APP, vehicle management backend, etc.). The pop-up clearly displays key information such as the time of the live object detection under the vehicle, the current driving position of the vehicle, and the preset category of the live object, so that the driver can be informed and take action in a timely manner after the driving ends.

[0120] 3. When the vehicle is locked and stationary, a warning response strategy is executed based on the preset categories of living objects under the vehicle and the response rules corresponding to the duration of their stay.

[0121] Specifically, when the system detects a living body under the vehicle and the vehicle is locked and stationary, the system strictly follows the tiered early warning response strategy based on the preset categories and duration of the living body under the vehicle, and accurately executes corresponding operations such as local audible and visual warnings, terminal pop-up warnings, and vehicle start restrictions, ensuring that there are no omissions or over-excessive safety controls under the vehicle during the stationary period.

[0122] In the above implementation, differentiated early warning and control schemes are formulated based on different vehicle operating states. When the vehicle is waiting to be started, the start-up process is restricted to avoid the dual risks of live body injury under the vehicle and vehicle mechanical failure. When the vehicle is in motion, the response level is reduced to completely avoid the problem of control operations distracting the driver's attention and causing driving safety hazards. At the same time, when the vehicle is locked and stationary, the graded early warning and control are strictly implemented to ensure the comprehensiveness of under-vehicle safety control. This avoids the problems of insufficient adaptability and unreasonable operation of single control rules in different vehicle states, improves the scenario adaptability and operational safety of the under-vehicle live body detection system, and takes into account the multiple needs of under-vehicle live body protection, vehicle equipment safety and road driving safety.

[0123] Furthermore, when the system performs any of the related operations such as under-vehicle liveness detection, graded warnings, one-click removal, and vehicle start restriction, it will simultaneously and automatically mark the behavior as an event. The marked content includes key information such as the event trigger time, the preset category of the liveness under the vehicle, the duration of the liveness in the effective detection area under the vehicle, the actual type of warning triggered, whether the one-click removal operation was performed and the removal result, and the triggering and lifting time of the vehicle start restriction. All marked events are systematically organized in chronological order, and vehicle owners can query historical marked events at any time through the vehicle's in-vehicle terminal or the bound terminal, which facilitates subsequent tracing, anomaly analysis, and situation review of under-vehicle liveness detection related behaviors.

[0124] Furthermore, for all information related to warning events generated during the vehicle undercarriage liveness detection process, including various core data triggered by the warning, the location information of the liveness under the vehicle, vehicle operating status information, and corresponding tagged event content, the system will perform high-strength encryption storage processing. After encrypting the data using a vehicle-specific encryption algorithm, it will be synchronously stored in the vehicle's local secure storage module and the encrypted cloud space of the bound terminal. The encrypted stored information will only be accessible to authorized users bound to the vehicle, and the data transmission process will use an encrypted transmission protocol to effectively prevent the illegal theft, tampering, or leakage of information related to warning events, ensuring vehicle safety and user data privacy. At the same time, the encrypted stored information will be retained according to a preset period to meet the user's long-term traceability needs.

[0125] In the above embodiments, by acquiring an image of the undercarriage when the vehicle is powered off as a reference image, comparing the reference image with a real-time image of the undercarriage to determine the image to be detected containing the difference region, extracting the feature detection information of the image to be detected, and thereby identifying and / or determining the status of the undercarriage liveness, it is possible to provide an accurate image reference that fits the actual use scenario of the vehicle for undercarriage liveness detection, effectively filtering out redundant image information without difference, accurately locking the suspected liveness region, improving the targeting and efficiency of undercarriage detection, and at the same time, relying on the feature detection information to achieve accurate identification and / or status determination of the undercarriage liveness, ensuring the accuracy and effectiveness of the undercarriage liveness detection results.

[0126] This application also provides an electronic device, such as... Figure 7 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0127] The electronic device may include components such as a processor 301 with one or more processing cores, a memory 302 with one or more storage media, a power supply 303, and an input unit 304. Those skilled in the art will understand that... Figure 6 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0128] The processor 301 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines, and performs various functions and processes data by running or executing computer programs and / or modules stored in the memory 302, and by calling data stored in the memory 302. Optionally, the processor 301 may include one or more processing cores; optionally, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 301.

[0129] The memory 302 can be used to store computer programs and modules. The processor 301 executes various functional applications and vehicle control by running the computer programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one computer program required for a function (such as voltage control), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 302 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include memory electronics to provide the processor 301 with access to the memory 302.

[0130] The electronic device also includes a power supply 303 that supplies power to the various components. Optionally, the power supply 303 can be logically connected to the processor 301 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 303 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0131] The electronic device may also include an input unit 304, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0132] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to the processes of one or more computer programs into the memory 302 according to the following instructions, and the processor 301 runs the computer programs stored in the memory 302 to realize various functions, such as:

[0133] When the vehicle is powered off, the image under the vehicle is used as the reference image;

[0134] Based on the reference image and the real-time image of the vehicle underside, the real-time image of the region with differences is identified as the image to be detected;

[0135] The feature detection information of the image to be detected is determined, and the identification of a living body under the vehicle and / or the determination of the state of the living body under the vehicle are realized based on the feature detection information.

[0136] Therefore, the electronic device provided in this application embodiment, by acquiring an image of the undercarriage as a reference image when the vehicle is powered off, comparing the reference image with a real-time image of the undercarriage to determine the image to be detected containing the difference region, extracting the feature detection information of the image to be detected, and thereby identifying and / or determining the status of the undercarriage liveness, can provide a precise image reference that fits the actual usage scenario of the vehicle for undercarriage liveness detection, effectively filtering redundant image information without difference, accurately locking the suspected liveness region, improving the targeting and efficiency of undercarriage detection, and at the same time, relying on the feature detection information to achieve accurate identification and / or status determination of the undercarriage liveness, ensuring the accuracy and effectiveness of the undercarriage liveness detection results.

[0137] For details on the specific implementation methods and corresponding beneficial effects of each of the above operations, please refer to the detailed description of the liveness detection method above, which will not be repeated here.

[0138] Therefore, this application provides a computer program product that stores instructions that, when executed by a computer, cause the computer to perform any of the liveness detection methods described in the above embodiments.

[0139] This application also provides a vehicle that includes the aforementioned electronic device or processor, the processor being used in real time for the liveness detection method in any of the above embodiments.

[0140] The above implementation method, by acquiring an image of the undercarriage when the vehicle is powered off as a reference image, comparing the reference image with a real-time image of the undercarriage to determine the image to be detected containing the difference region, extracting the feature detection information of the image to be detected, and thereby identifying and / or determining the status of the undercarriage liveness, can provide a precise image reference that fits the actual usage scenario of the vehicle for undercarriage liveness detection, effectively filtering out redundant image information without difference, accurately locking the suspected liveness region, improving the targeting and efficiency of undercarriage detection, and at the same time, relying on feature detection information to achieve accurate identification and / or status determination of the undercarriage liveness, ensuring the accuracy and effectiveness of the undercarriage liveness detection results.

[0141] The foregoing has provided a detailed description of a vehicle underbody liveness detection method, computer program product, electronic device, and vehicle provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only configured to help understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0142] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for detecting liveness under a vehicle, characterized in that, The detection method includes: When the vehicle is powered off, the image under the vehicle is used as the reference image; Based on the reference image and the real-time image of the vehicle underside, the real-time image of the region with differences is identified as the image to be detected; The feature detection information of the image to be detected is determined, and the identification of a living body under the vehicle and / or the determination of the state of the living body under the vehicle are realized based on the feature detection information.

2. The detection method according to claim 1, characterized in that, The feature detection information includes at least the category, location information, and confidence level information of the target under the vehicle. The process of identifying a live object under the vehicle and / or determining the state of the live object based on the feature detection information includes: The identification of the living body under the vehicle is completed based on the category of the target to be detected, the location information, and the confidence information. When it is determined that there is a living body under the vehicle, the state of the living body under the vehicle is determined according to the change in the position information, the state including entering the vehicle or leaving the vehicle.

3. The detection method according to claim 2, characterized in that, The detection method further includes: When it is determined that a living body exists under the vehicle, the living body is marked. Alternatively, record the time when the living person under the vehicle leaves the vehicle and / or enters the vehicle; Alternatively, a preset early warning response strategy can be triggered to encrypt and store information related to the early warning event.

4. The detection method according to claim 3, characterized in that, The early warning response strategy is determined based on the type and duration of the living object under the vehicle. When a living object under the vehicle is classified as the first preset category and its dwell time is less than a preset threshold, the vehicle's audio-visual warning is triggered. When a living object under the vehicle is classified as the first preset category and its dwell time exceeds a preset threshold, or is classified as the second preset category, a local audible and visual warning is triggered and a pop-up warning is sent to the terminal bound to the vehicle. When the living object under the vehicle is classified as the third preset category, a pop-up warning is sent to the terminal bound to the vehicle, and the vehicle is restricted from starting until the living object under the vehicle leaves.

5. The detection method according to claim 3, characterized in that, The early warning response strategy is adjusted based on the current state of the vehicle; When the vehicle is in a ready-to-start state, the vehicle's start-up is restricted; When the vehicle is in motion, the response level is reduced and a pop-up warning is sent to the terminal bound to the vehicle. When the vehicle is locked and stationary, the warning response strategy is executed based on the preset category of the living body under the vehicle and the response rules corresponding to the dwell time.

6. The detection method according to claim 1, characterized in that, The step of determining the real-time image with discrepancies based on the reference image and the real-time image of the vehicle underside as the image to be detected includes: The lighting environment of the real-time image is determined based on image brightness feature parameters; If the real-time image is from a low-light environment, perform image enhancement processing on the real-time image; Based on the real-time image after image enhancement processing, the real-time image under bright light conditions without processing, and the reference image, the real-time image with the difference region is determined as the image to be detected.

7. The detection method according to claim 1, characterized in that, The step of determining the real-time image with discrepancies based on the reference image and the real-time image of the vehicle underside as the image to be detected includes: The absolute difference image is generated by subtracting the reference image from the real-time image pixel by pixel and taking the absolute value. The absolute difference image is converted into a binary image by using a preset grayscale threshold; The binary image is processed to obtain a valid connected region, and the real-time image containing the valid connected region is determined as the image to be detected.

8. A computer program product, characterized in that, The computer program product stores instructions that, when executed by a computer, cause the computer to perform the liveness detection method according to any one of claims 1-7.

9. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the liveness detection method according to any one of claims 1-7.

10. A vehicle, characterized in that, The vehicles include: The electronic device as described in claim 9; Alternatively, a processor, said processor being configured to perform the liveness detection method according to any one of claims 1-7.

Images