Electric tailgate anti-trap control method, apparatus, system, and vehicle

Abstract

This invention provides a method, device, system, and vehicle for anti-pinch control of an electric tailgate, relating to the field of vehicle control technology. The method includes: receiving raw radar point cloud data generated by a millimeter-wave radar sensor in an obstacle sensing module; receiving tailgate motion data sensed by a tailgate motion sensor in an electric tailgate actuator; filtering the raw radar point cloud data to generate preprocessed data; identifying and classifying obstacles in the preprocessed data using an obstacle recognition and classification model to determine the obstacle type; determining the position, motion data, and distance of the obstacle based on the preprocessed data; determining a critical distance based on the obstacle's motion data, tailgate motion data, and a predefined braking response time; determining an anti-pinch control command based on the obstacle type, obstacle distance, and critical distance; and sending the anti-pinch control command to the electric tailgate actuator. This invention improves the sensitivity of anti-pinch control and enhances safety.

Description

Technical Field

[0001] This invention relates generally to the field of vehicle control technology, and more specifically to an anti-pinch control method, device, system, and vehicle for electric tailgates. Background Technology

[0002] Current electric tailgate anti-pinch technologies for vehicles are mainly divided into three categories: mechanical contact anti-pinch (such as anti-pinch strips), infrared photoelectric anti-pinch, and ultrasonic anti-pinch.

[0003] Mechanical contact anti-pinch relies on physical structures such as anti-pinch strips, and braking can only be triggered after contact with an obstacle. This results in a "collision before braking" lag problem, low anti-pinch sensitivity, and the anti-pinch strips are prone to failure due to aging and deformation, providing poor protection for children's fingers, slender objects, and other small components.

[0004] Infrared photoelectric anti-pinch technology detects obstacles by blocking infrared light, but it is greatly affected by the lighting environment. It is prone to misjudgment (such as false braking or missed detection of obstacles) in rainy, foggy, strong light, or dusty conditions. In addition, the detection range is limited to the light propagation path, and there are obvious detection blind spots.

[0005] Ultrasonic anti-pinch devices use the principle of ultrasonic reflection to detect obstacles, but their accuracy in recognizing non-metallic objects is insufficient. They are easily affected by environmental noise such as vehicle vibration and external sound waves. The anti-pinch response delay is usually 100-200ms, and there is a "blind zone" when detecting at close range, which cannot cover the entire area of ​​the tailgate's movement trajectory.

[0006] While anti-pinch technology for electric tailgates has become a mainstream feature, existing anti-pinch technologies for electric tailgates have drawbacks such as "contact-based triggering lag," "non-contact-based technologies are prone to misjudgment and have blind spots," and "slow response speed." These technologies also have problems in multiple dimensions, including environmental adaptability, perception and recognition accuracy, and hardware compatibility. These issues not only affect the user experience but may also pose safety hazards. Summary of the Invention

[0007] To address the aforementioned problems in the prior art, in a first aspect, embodiments of the present invention provide an anti-pinch control method for an electric tailgate. The method includes: receiving raw radar point cloud data associated with an obstacle generated by a millimeter-wave radar sensor in an obstacle sensing module; receiving tailgate motion data sensed by a tailgate motion sensor in an electric tailgate actuator; filtering the raw radar point cloud data to generate preprocessed data; identifying and classifying obstacles in the preprocessed data using an obstacle recognition and classification model to determine the obstacle type; determining the position, motion data, and distance of the obstacle based on the preprocessed data; determining a critical distance based on the obstacle's motion data, tailgate motion data, and a predefined braking response time; determining an anti-pinch control command based on the obstacle type, the obstacle's distance, and the critical distance; and sending the anti-pinch control command to the electric tailgate actuator.

[0008] In some embodiments, the tailgate motion data includes the tailgate's speed and acceleration, and the obstacle motion data includes the obstacle's moving speed. Furthermore, determining the critical distance based on the obstacle's motion data, the tailgate motion data, and a predefined braking response time includes: determining the critical distance based on the tailgate's speed, tailgate acceleration, obstacle's moving speed, and braking response time, wherein the critical distance is positively correlated with the tailgate's speed, tailgate acceleration, obstacle's moving speed, and braking response time.

[0009] In some implementations, the obstacle type includes one or more of dynamic flexible obstacles, dynamic rigid obstacles, static rigid obstacles, and non-hazardous targets.

[0010] In some implementations, determining the anti-pinch control command based on the obstacle type, the distance to the obstacle, and the critical distance includes: for dynamic flexible obstacles, if the distance to the obstacle is less than the critical distance, the anti-pinch control command is determined to control the tailgate motor to stop and then control the motor to rotate in the opposite direction by a predetermined angle; for dynamic rigid obstacles and static rigid obstacles, if the distance to the obstacle is less than the critical distance, the anti-pinch control command is determined to control the tailgate motor to stop; for non-dangerous targets, no anti-pinch control command is generated.

[0011] In some implementations, the obstacle identification and classification of the preprocessed data using an obstacle identification and classification model to determine the obstacle type includes: extracting features of the obstacle at multiple different scales from the preprocessed data through the feature extraction layer in the obstacle identification and classification model to generate feature maps at multiple different scales; calculating the category probability distribution of the predicted anchor boxes on the feature maps through the classification prediction head in the obstacle identification and classification model, and determining the obstacle type based on the category probability distribution.

[0012] In some implementations, the obstacle recognition classification model is obtained through the following steps: constructing an electric tailgate anti-pinch scenario dataset, wherein the samples in the electric tailgate anti-pinch scenario dataset include radar feature maps under various anti-pinch scenarios and the true categories of labeled samples; inputting the electric tailgate anti-pinch scenario dataset into the obstacle recognition classification model, and outputting a predicted category through the obstacle recognition classification model; generating a loss function based on the difference between the predicted category and the true category, as well as the category weight and / or feature scale weight, and training the obstacle recognition classification model using the loss function.

[0013] In a second aspect, embodiments of the present invention provide an anti-pinch control device for an electric tailgate, the device including a memory and a processor, the memory storing a computer program that, when executed by the processor, implements the anti-pinch control method for an electric tailgate as described in any of the above embodiments.

[0014] In a third aspect, embodiments of the present invention provide an anti-pinch control system for an electric tailgate, the system comprising an obstacle sensing module, an electric tailgate actuator, and an electric tailgate anti-pinch control device as described in any of the above embodiments. The obstacle sensing module includes a millimeter-wave radar sensor disposed at a position relevant to the movement of the electric tailgate of the vehicle, for sensing obstacles in the electric tailgate's movement environment and generating raw radar point cloud data associated with the obstacles. The electric tailgate actuator includes a tailgate motor and a tailgate motion sensor, wherein the tailgate motion sensor is used to sense tailgate movement data, and the tailgate motor is used to execute anti-pinch control commands.

[0015] In some embodiments, the obstacle sensing module includes at least three millimeter-wave radar sensors, which are respectively arranged at the center of the upper edge and the two side edges of the inner side of the power tailgate.

[0016] In a fourth aspect, embodiments of the present invention provide a vehicle that includes the electric tailgate anti-pinch control system described in any of the above embodiments.

[0017] To address the shortcomings of existing electric tailgate anti-pinch technologies, such as "contact-based trigger lag," "non-contact-based misjudgment and blind spots," and "slow response speed," this invention provides a non-contact anti-pinch solution based on millimeter-wave radar. This solution enables early obstacle identification without physical contact, improves anti-pinch sensitivity and recognition accuracy, and reduces false and false detection rates. Simultaneously, it simplifies hardware installation complexity, adapts to the electric tailgate structures of different vehicle models, and ensures the safety of personnel and goods.

[0018] The electric tailgate anti-pinch control scheme proposed in the embodiments of the present invention can be applied to various vehicles equipped with electric tailgates, such as passenger cars and commercial vehicles. It can be integrated into the original electric tailgate system or aftermarket electric tailgate modification kit, covering vehicle usage scenarios such as family, business, and logistics transportation, and has a wide range of application scenarios. Attached Figure Description

[0019] The above and other objects, features, and advantages of embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0020] Figure 1 A flowchart of an electric tailgate anti-pinch control method according to an embodiment of the present invention is shown;

[0021] Figure 2 A schematic diagram of an example layout of a millimeter-wave radar sensor according to an embodiment of the present invention is shown.

[0022] Figure 3 A schematic block diagram of an electric tailgate anti-pinch control system according to an embodiment of the present invention is shown.

[0023] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts. Detailed Implementation

[0024] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are given merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way.

[0025] In one aspect, embodiments of the present invention provide a method for preventing pinching of an electric tailgate. (See reference...) Figure 1 The diagram shows a flowchart of an electric tailgate anti-pinch control method 100 according to an embodiment of the present invention. The method 100 includes steps S101-S108.

[0026] In step S101, raw radar point cloud data associated with obstacles, generated by the millimeter-wave radar sensor in the obstacle sensing module, is received. The millimeter-wave radar sensor collects real-time obstacle data (such as the position and distance of people and objects) of the tailgate and surrounding area.

[0027] As an example only, a millimeter-wave radar sensor can be a 77GHz band millimeter-wave radar sensor, such as the model XX-7701, which has a detection range of 10cm-5m, an angular resolution of ≤1°, and a distance resolution of ≤5cm, and is used to accurately capture the position, speed, and outline information of obstacles.

[0028] The radar data acquisition process can occur, for example, when the electric tailgate is opened / closed. The millimeter-wave radar sensor synchronously acquires raw obstacle data (point cloud, distance, motion status, etc.) within the range of the tailgate's movement trajectory, and the sampling frequency can be, for example, 100Hz.

[0029] In one embodiment of the present invention, the obstacle sensing module includes at least three millimeter-wave radar sensors. (See reference...) Figure 2 The diagram illustrates an example layout of a millimeter-wave radar sensor according to an embodiment of the present invention. The diagram includes three millimeter-wave radar sensors: a left millimeter-wave radar sensor, a middle millimeter-wave radar sensor, and a right millimeter-wave radar sensor, respectively arranged along the middle of the upper edge of the inner side of the electric tailgate and the two side edges, forming a triangular detection layout. This layout can cover the entire detection range of the tailgate's movement trajectory, eliminating the detection blind spots that are easily caused by a single radar.

[0030] However, it should be noted that the implementation of the present invention is not limited to setting three sensors, as long as the sensing range of the sensors can cover the tailgate movement area.

[0031] Optionally, the obstacle sensing module may also include a data processing unit for performing preliminary data processing on the raw radar data sensed by the millimeter-wave radar sensor, such as preliminary data filtering and obstacle recognition. As an example only, the data processing unit may employ an ARM Cortex-A53 processor.

[0032] In step S102, tailgate motion data sensed by the tailgate motion sensor in the electric tailgate actuator is received. The tailgate motion sensor may, for example, include a Hall sensor.

[0033] In step S103, the raw radar point cloud data is filtered to generate preprocessed data. This step can be multi-dimensional data processing, which involves noise reduction of the raw radar data to remove interference such as vehicle vibration and environmental noise, while retaining effective obstacle features to prepare for subsequent identification.

[0034] As one embodiment of the present invention, filtering the raw radar point cloud data may include performing Gaussian filtering and median filtering on the raw radar point cloud data. For example, Gaussian filtering may use a filtering window of 3×3, and median filtering may use a sorted window of 5 data points to filter, thereby removing interference data such as vehicle vibration and environmental noise, and retaining effective obstacle feature information.

[0035] Gaussian filtering primarily filters high-frequency random noise. Based on the weighting logic of the Gaussian function, it performs a weighted average of pixel / point cloud values ​​within the neighborhood of radar data, thus smoothing the data and suppressing high-frequency interference. Specific information filtered may include environmental electromagnetic noise, high-frequency jitter of the radar signal itself, and short-term environmental interference (such as momentary obstruction from dust or raindrops).

[0036] Median filtering primarily filters impulse noise and outliers. It sorts the pixel / point cloud values ​​in the neighborhood of radar data and takes the median value as the output. Its core advantage is that it can efficiently remove "outliers that are extremely different from the surrounding data". Specific information filtered can include, for example, impulse interference caused by vehicle vibration, instantaneous strong reflection interference from non-target objects, and clutter from distant irrelevant targets.

[0037] The above steps employ a highly reliable data acquisition scheme using multi-radar collaboration and combined filtering. For example, three millimeter-wave radars are used to simultaneously acquire data (100Hz sampling frequency) to cover the entire movement trajectory of the tailgate without blind spots. At the same time, a combination of Gaussian filtering and median filtering is used for preprocessing to specifically eliminate interference from vehicle vibration, environmental noise, and other factors.

[0038] In step S104, the preprocessed data is used to identify and classify obstacles using an obstacle recognition and classification model to determine the type of obstacle.

[0039] Obstacle recognition and classification models, such as machine learning models, can be derived by improving upon the YOLO-Lite algorithm. For example, such models can analyze preprocessed data to identify obstacle types such as pedestrians, limbs, and objects; distinguish between static obstacles (such as walls) and dynamic obstacles (such as approaching hands or pets); and exclude non-dangerous targets (such as fixed structures along the normal movement path of the tailgate, or soft items inside the vehicle such as floor mats or sponges), thus avoiding misjudgments of non-dangerous targets.

[0040] In step S105, the position, motion data, and distance of the obstacle are determined based on the preprocessed data.

[0041] In step S106, the critical distance is determined based on the obstacle's motion data, the tailgate's motion data, and the predefined braking response time.

[0042] In one embodiment of the present invention, the tailgate motion data includes the tailgate speed and acceleration, and the obstacle motion data includes the obstacle's moving speed. Step S106 may include determining a critical distance based on the tailgate speed, tailgate acceleration, obstacle's moving speed, and braking response time, wherein the critical distance is positively correlated with the tailgate speed, tailgate acceleration, obstacle's moving speed, and braking response time.

[0043] As a specific example, the critical distance can be determined according to the following formula: d = v × t + 0.5 × a × t² + v_obst × t, where d is the critical distance, v is the speed of the tailgate, t is the braking response time, a is the acceleration of the tailgate, and v_obst is the moving speed of the obstacle.

[0044] In the example above, a dynamic safety distance model was established based on the current movement speed of the tailgate (e.g., adjustable from 0.1 to 0.5 m / s), the tailgate speed (v), acceleration (a), and obstacle movement speed (v_obst) fed back by the Hall sensor. The critical distance d was calculated in real time using a formula. The critical distance for triggering the anti-pinch brake was calculated in real time; the faster the speed, the larger the critical distance.

[0045] As an example, the predefined braking response time can be a value in the range of 30-80ms, for example, it can be 50ms.

[0046] ISO 26262 (Road Vehicle Functional Safety Standard) requires that the "ASIL-B level" function (such as the electric tailgate anti-pinch function) must have a "perception-decision-execution" end-to-end latency of ≤100ms for safety-related electronic control systems. The implementation of this invention limits the braking response time (from "detecting an obstacle entering the critical distance" to "the actuator starting to brake") to 50ms, reserving sufficient redundancy (the remaining 50ms can cover potential delays such as sensor data transmission and control module computing power fluctuations), ensuring that the system can still meet safety standards under extreme conditions.

[0047] Compared with mainstream vehicle safety functions (such as anti-pinch windows and sub-functions of automatic emergency braking AEB), the delay of critical response nodes is controlled within 30-80ms. 50ms can balance response speed and system stability. Too short a delay may cause false triggering due to signal jitter, while too long a delay may not provide timely protection.

[0048] The embodiments of this invention employ real-time calculation logic and a dynamic adaptation scheme for dynamic safety distance. A dynamic critical distance formula is constructed based on the tailgate speed (v), acceleration (a), and obstacle movement speed (v_obst), and real-time calculation is implemented. The critical distance is adjusted in real time according to the movement state of the tailgate and the obstacle, with a response time, for example, ≤50ms, meeting automotive-grade safety requirements.

[0049] In step S107, the anti-pinch control command is determined based on the obstacle type, the distance to the obstacle, and the critical distance.

[0050] Step S107 performs anti-pinch judgment, such as establishing a mapping table between obstacle type and critical distance, and judging whether the anti-pinch trigger condition is met based on the current distance of the obstacle and the critical distance. For example, the trigger condition is: the actual distance of the obstacle ≤ the critical distance d.

[0051] As one embodiment of the present invention, the obstacle type includes one or more of the following: dynamic flexible obstacle, dynamic rigid obstacle, static rigid obstacle, and non-dangerous target.

[0052] As one embodiment of the present invention, step S107 may include: for dynamic flexible obstacles, when the distance to the obstacle is less than the critical distance, the anti-pinch control command is determined to control the tailgate motor to stop and then control the motor to rotate in the opposite direction by a predetermined angle; for dynamic hard obstacles and static hard obstacles, when the distance to the obstacle is less than the critical distance, the anti-pinch control command is determined to control the tailgate motor to stop; for non-dangerous targets, no anti-pinch control command is generated.

[0053] As a specific example, Table 1 below illustrates differentiated anti-pinch strategies based on obstacle type.

[0054] Table 1. Specific Examples of Differentiated Anti-Pinch Strategies Based on Obstacle Type

[0055] Obstacle types Risk level Anti-pinch strategy Dynamic flexible barriers (core protection targets: pedestrians / limbs, such as hands, arms, and children's body parts) Highest Based on the dynamic safety distance formula (d=v×t+0.5×a×t²+v_obst×t), due to the dynamic movement of the target (v_obst>0), the calculated critical distance d will be farther than that of a static target, allowing for more braking buffer time. When the anti-pinch function is triggered, a "mild braking + slight reverse" action is performed—first, the motor power is cut off, and then the motor is controlled to rotate in the reverse direction by ≤5° to avoid impact on the human body from hard braking (such as preventing additional compression due to sudden stopping after a finger is pinched). Dynamic rigid obstacles (such as moving suitcases, backpacks, and trolleys) Medium and high d is calculated normally according to the formula. Since v_obst > 0, d is slightly larger than the static hard target to ensure timely braking. When the anti-pinch function is triggered, only the "quick stop" action is performed - directly cutting off the motor power supply without reverse rotation (to avoid hard objects colliding with other parts of the tailgate due to reverse movement), thus balancing protection and mechanical protection. Static rigid obstacles (such as stationary walls, pillars, and fixed furniture) medium The formula d is calculated normally. Since v_obst=0, the formula simplifies to d=v×t+0.5×a×t². The critical distance d is smaller than the dynamic target (no additional movement buffer is needed). When the anti-pinch function is triggered, the "instant stop" action is executed—the motor power is quickly cut off to prevent the tailgate from continuously colliding with obstacles (such as preventing the tailgate from hitting a wall and causing the strut to deform). Non-hazardous targets (such as tailgate fixing structures, soft interior trim, hinges, floor mats, and foam). Low The anti-pinch function is not triggered.

[0056] The embodiments of this invention implement a differentiated braking control strategy based on target type. When the anti-pinch mechanism is triggered, it abandons the traditional "one-size-fits-all" braking logic and adjusts the execution action according to the type of obstacle. For dynamic and flexible targets such as the human body and limbs, it avoids secondary impacts through gentle braking, such as "cutting off the power + reverse rotation ≤5°", thus ensuring safety and improving the experience. For soft materials inside the vehicle, such as sponges, a no-braking strategy is adopted to enhance the customer experience.

[0057] The embodiment of the present invention adjusts the anti-pinch critical distance in real time according to the tailgate movement state and obstacle conditions, which is different from the fixed safety distance and takes into account both sensitivity and stability.

[0058] In step S108, an anti-pinch control command is sent to the electric tailgate actuator.

[0059] The electric tailgate actuator is the execution end. After receiving the anti-pinch control command, it completes actions such as opening, closing, stopping, or reversing the tailgate to achieve anti-pinch braking. When an obstacle is detected entering the critical distance, the method outputs a braking signal to the electric tailgate controller and adjusts the braking force according to the type of obstacle. For dynamic flexible obstacles (such as a human body), it uses gentle braking to avoid secondary impact.

[0060] As an example only, after the anti-pinch brake signal is output, if it is determined that anti-pinch is needed, the controller immediately cuts off the power to the drive motor and simultaneously controls the motor to rotate slightly in the opposite direction (i.e., the tailgate rises) (e.g., the angle is ≤5°) to complete the anti-pinch action. If no obstacle is detected, or it is determined that anti-pinch is not needed, the tailgate will operate normally at a preset speed until it is fully open or closed.

[0061] The following is a detailed introduction to obstacle recognition and classification models.

[0062] As one embodiment of the present invention, the obstacle identification and classification of preprocessed data by an obstacle identification and classification model to determine the obstacle type may include: extracting features of the obstacle at multiple different scales from the preprocessed data through the feature extraction layer in the obstacle identification and classification model to generate feature maps at multiple different scales; calculating the category probability distribution of the prediction anchor boxes on the feature maps through the classification prediction head in the obstacle identification and classification model, and determining the obstacle type based on the category probability distribution.

[0063] As one embodiment of the present invention, the obstacle recognition classification model is obtained through the following steps: constructing an electric tailgate anti-pinch scenario dataset, wherein the samples in the electric tailgate anti-pinch scenario dataset include radar feature maps under various anti-pinch scenarios and the true categories of the labeled samples; inputting the electric tailgate anti-pinch scenario dataset into the obstacle recognition classification model, and outputting the predicted category through the obstacle recognition classification model; generating a loss function based on the difference between the predicted category and the true category, as well as the category weight and / or feature scale weight, and training the obstacle recognition classification model using the loss function.

[0064] The following explanation uses an obstacle recognition and classification model built on the YOLO-Lite algorithm as an example.

[0065] The YOLO-Lite algorithm is a lightweight object detection algorithm that, at its core, directly outputs the "category probability" and "location coordinates" of obstacles through an "end-to-end" network structure. It is suitable for embedded devices and boasts advantages such as fast recognition speed and low resource consumption. The implementation of this invention optimizes its feature extraction network to improve the recognition capability of small targets. This invention's implementation addresses the classification needs (pedestrians / limbs / objects / non-dangerous targets) in electric tailgate scenarios, achieving classification functionality through both network structure design and training optimization.

[0066] First, in terms of network structure, classification results are output through "feature extraction + multi-scale detection head".

[0067] YOLO-Lite's classification ability relies on the collaboration between the "convolutional feature extraction layer" and the "classification prediction head". The specific process includes: the feature extraction layer extracts key features of obstacles; the classification prediction head outputs class probabilities.

[0068] In the process of extracting key features of obstacles in the feature extraction layer, the original radar point cloud is preprocessed by filtering and then converted into a three-dimensional feature map of "distance-angle-velocity" before being input into the YOLO-Lite feature extraction network. The implementation of this invention can optimize the original network by replacing some ordinary convolutions with depthwise separable convolutions, maintaining lightweight design while improving feature extraction capabilities. The feature extraction layer gradually extracts "local detail features" (such as the outline of limbs and the shape of objects) and "global semantic features" (such as the overall posture of a human body and the movement trend of objects) of obstacles through multiple sets of convolutions (including ReLU activation functions), ultimately outputting multiple (e.g., three) feature maps of different scales (e.g., 13×13, 26×26, 52×52), corresponding to the feature representations of large targets such as a complete human body, medium targets such as a box, and small targets such as a finger, respectively.

[0069] During the process of outputting category probabilities by the classification prediction head, each scale of feature map is connected to a "classification prediction head" (composed of a 1×1 convolutional layer). This prediction head calculates the "category probability distribution" for each "prediction anchor box" (a pre-defined frame with different aspect ratios to adapt to the obstacle size in the electric tailgate scenario, such as a 10×20 anchor box for fingers and a 50×50 anchor box for boxes) on the feature map. According to the anti-pinch requirements, the classification target can be defined as four categories: pedestrians (e.g., children, adults), limbs (e.g., hands, arms), objects (e.g., suitcases, backpacks), and non-dangerous targets (e.g., tailgate hinges, soft mats). The classification prediction head outputs the probability of each anchor box belonging to these four categories through the Softmax function. The category with the highest probability is the final classification result for the obstacle. For example, if an anchor box outputs a "limb" probability of 0.92 and other category probabilities < 0.1, it is determined to be a limb.

[0070] In terms of training optimization, the implementation method of this invention improves classification accuracy based on the electric tailgate scene dataset.

[0071] To ensure that the classification results are suitable for in-vehicle anti-pinch scenarios (such as avoiding misclassifying an "arm" as an "object" or a "floor mat" as an "obstacle"), the implementation of this invention can improve classification accuracy through the construction of scenario-based datasets and optimization of loss functions:

[0072] 1) Construct an “Electric Tailgate Anti-Pinch Scene Dataset”: This dataset contains radar feature maps of different lighting conditions (day / night), environments (rain / dust), and obstacle types (1000+ samples, such as fingers at different angles, boxes of different sizes, and soft objects of different materials). Each sample is labeled with its “true category” and “location” to ensure that the dataset covers all possible targets in the anti-pinch scene.

[0073] 2) Optimize the classification loss function: The “cross-entropy loss function” is used to calculate the difference between the predicted category and the true category. At the same time, “category weight” is introduced (high-risk categories such as “limbs” and “pedestrians” are given higher weights, such as limbs weight 1.5 and non-dangerous targets weight 0.5), which forces the network to pay more attention to the classification accuracy of high-risk targets during training, and finally achieves a classification accuracy of ≥95% (non-dangerous target misclassification rate <3%).

[0074] Due to its lightweight design, the original YOLO-Lite has a weak ability to recognize small targets (such as fingers or small parts in electric tailgate scenarios, objects typically <5cm in size) (prone to missed or false recognition). The implementation of this invention solves this problem through the following three improvements, ensuring a small target recognition recall rate of ≥90%.

[0075] Point 1, Multi-scale feature fusion: Enhance the feature representation of small targets.

[0076] Because small targets have a low pixel ratio in the original image / feature map (e.g., a finger only occupies 2×3 pixels in a 52×52 feature map), a single-scale feature is insufficient to fully express its details. The implementation of this invention can introduce a cross-scale feature fusion module (a simplified version of the FPN structure): The large-scale feature map (e.g., 13×13, representing global semantics) is "upsampled" (enlarged to 26×26, 52×52) and "element-by-elementally added" to the medium-scale (26×26) and small-scale (52×52) feature maps. This allows the small-scale feature map to simultaneously possess both "local detail features of the small target" and "global semantic features of the large target" (e.g., edge details of the finger + posture semantics of the human body), solving the problem of "insufficient feature information" for small targets. For example, the small-scale feature map (52×52) can only capture the local edge of the finger. After fusing with the large-scale feature, it can be combined with the "movement trend of the human arm" to assist in determining "whether the edge belongs to the finger," avoiding misjudging a "tailgate gap" as a "finger."

[0077] Secondly, adapt the anchor frame size: design a custom anchor frame for small targets.

[0078] The original YOLO-Lite anchor box sizes are usually designed based on general scenarios (such as road targets), which are not suitable for the small target sizes of electric tailgates (e.g., the minimum general anchor box size is 16×32, while the size of a finger in the feature map is only 10×20). The implementation of this invention can redesign the anchor boxes using the K-means clustering algorithm: K-means clustering is performed on the bounding boxes of all small targets (size < 5cm) in the "Electric Tailgate Anti-Pinch Scene Dataset" to obtain 3 sets of anchor box sizes that are suitable for small targets: such as 8×18 (corresponding to small components, such as pens), 12×22 (corresponding to fingers), and 15×25 (corresponding to small parts, such as keys), to ensure that the anchor boxes are highly matched with the real size of the small targets, reduce the initial error between the predicted box and the real box, and improve the detection recall rate of small targets (avoiding the missed detection of small targets due to anchor box mismatch).

[0079] Thirdly, optimize the detection head: enhance the detection weights of small-scale feature maps.

[0080] The original YOLO-Lite assigns consistent detection weights to feature maps across all three scales. However, small targets are primarily represented in the small-scale feature map (52×52). The implementation of this invention enhances small target detection by adjusting the detection head weights: when calculating the final detection loss, the detection loss of the small-scale feature map (52×52) is given higher weights (e.g., 1.2 for small scale, 0.9 for medium scale, and 0.8 for large scale). This forces the network to pay more attention to small targets on the small-scale feature map during training, improving its sensitivity to small components such as "fingers" and "small parts," and preventing the network from neglecting small targets due to "biased towards large target detection."

[0081] The implementation of this invention adopts a refined obstacle recognition and classification based on YOLO-Lite, which breaks through the traditional coarse logic of "only judging whether there is an obstacle". It achieves refined recognition through the improved YOLO-Lite algorithm - obstacle type (pedestrian / limb / object), dynamic / static attribute (static / dynamic), risk level (dangerous / non-dangerous). It can actively exclude non-dangerous targets such as fixed structures of tailgates and soft objects, so that the misjudgment rate of non-dangerous targets is <3%.

[0082] On the other hand, embodiments of the present invention provide an anti-pinch control device for electric tailgates, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, it implements the anti-pinch control method for electric tailgates described in any of the above embodiments.

[0083] On the other hand, embodiments of the present invention provide an anti-pinch control system for electric tailgates. (See reference...) Figure 3This diagram illustrates a schematic block diagram of an anti-pinch control system for a power tailgate according to an embodiment of the present invention. The system includes an obstacle sensing module, a power tailgate actuator, and the anti-pinch control device for the power tailgate described in the preceding embodiments. For example, the anti-pinch control device can communicate with the obstacle sensing module and the power tailgate actuator via a CAN bus, and the detection data from the millimeter-wave radar can be uniformly transmitted to the anti-pinch control device via the CAN bus.

[0084] The obstacle sensing module includes a millimeter-wave radar sensor, which is positioned at a location relevant to the movement of the vehicle's electric tailgate. The millimeter-wave radar sensor is used to sense obstacles in the electric tailgate's moving environment and generate raw radar point cloud data associated with the obstacles.

[0085] The radar data acquisition process can occur, for example, when the electric tailgate is opened / closed. The millimeter-wave radar sensor synchronously acquires raw obstacle data (point cloud, distance, motion status, etc.) within the range of the tailgate's movement trajectory, and the sampling frequency can be, for example, 100Hz.

[0086] In one embodiment of the present invention, the obstacle sensing module includes at least three millimeter-wave radar sensors. (See reference...) Figure 2 The diagram illustrates an example layout of a millimeter-wave radar sensor according to an embodiment of the present invention. The diagram includes three millimeter-wave radar sensors—left, center, and right—arranged along the middle and side edges of the upper inner edge of the power tailgate, forming a triangular detection layout. This layout covers the entire area of ​​the tailgate's movement trajectory, eliminating blind spots. The embodiment of the present invention achieves optimized selection and installation layout design of the millimeter-wave radar hardware.

[0087] However, it should be noted that the implementation of the present invention is not limited to setting three sensors, as long as the sensing range of the sensors can cover the tailgate movement area.

[0088] Optionally, the obstacle sensing module may also include a data processing unit for performing preliminary data processing on the raw radar data sensed by the millimeter-wave radar sensor, such as preliminary data filtering and obstacle recognition. As an example only, the data processing unit may employ an ARM Cortex-A53 processor.

[0089] The power tailgate actuator includes a tailgate motor and a tailgate motion sensor. The tailgate motion sensor senses tailgate movement data, and the tailgate motor executes anti-pinch control commands. The tailgate motion sensor may include, for example, a Hall effect sensor. The power tailgate actuator is the actuating end; upon receiving the anti-pinch control command, it performs actions such as opening, stopping, or reversing the tailgate to achieve anti-pinch braking.

[0090] In another aspect, embodiments of the present invention provide a vehicle that includes an electric tailgate anti-pinch control system as described in any of the above embodiments.

[0091] In another aspect, embodiments of the present invention also provide a computer program product including computer-readable instructions that, when executed by a processor, perform the steps of the electric tailgate anti-pinch control method described in any of the above embodiments.

[0092] In another aspect, embodiments of the present invention also provide a storage medium storing computer-readable instructions that, when executed by a processor, perform the electric tailgate anti-pinch control method described in any of the above embodiments.

[0093] To address the shortcomings of existing electric tailgate anti-pinch technologies, such as "contact-based trigger lag," "non-contact-based misjudgment and blind spots," and "slow response speed," this invention provides a non-contact anti-pinch solution based on millimeter-wave radar. This solution enables early obstacle identification without physical contact, improves anti-pinch sensitivity and recognition accuracy, and reduces false and false detection rates. Simultaneously, it simplifies hardware installation complexity, adapts to the electric tailgate structures of different vehicle models, and ensures the safety of personnel and goods.

[0094] The electric tailgate anti-pinch control scheme proposed in the embodiments of the present invention can be applied to various vehicles equipped with electric tailgates, such as passenger cars and commercial vehicles. It can be integrated into the original electric tailgate system or aftermarket electric tailgate modification kit, covering vehicle usage scenarios such as family, business, and logistics transportation, and has a wide range of application scenarios.

[0095] The foregoing description of embodiments of the invention has been given for illustrative purposes and is not exhaustive, nor is it intended to limit the invention to the exact forms disclosed. Those skilled in the art will understand that various changes can be made without departing from the scope of the invention, and elements therein can be substituted with equivalents. Furthermore, many modifications can be made to adapt particular situations or materials to the teachings of the invention without departing from the basic scope of the invention. Therefore, the invention is not intended to be limited to the specific embodiments disclosed as the best mode contemplated for carrying out the invention; the invention will include all embodiments falling within the scope of the appended claims.

Claims

1. A method for preventing pinching of an electric tailgate, characterized in that, The method includes: Receive raw radar point cloud data associated with obstacles generated by the millimeter-wave radar sensor in the obstacle sensing module; Receive tailgate motion data sensed by the tailgate motion sensor in the electric tailgate actuator; The raw radar point cloud data is filtered to generate preprocessed data; The preprocessed data is used to identify and classify obstacles using an obstacle recognition and classification model to determine the obstacle type. The obstacle type includes one or more of the following: dynamic flexible obstacles, dynamic rigid obstacles, static rigid obstacles, and non-dangerous targets. The location, motion data, and distance of the obstacle are determined based on the preprocessed data; The critical distance is determined based on the obstacle's motion data, the tailgate's motion data, and the predefined braking response time; The anti-pinch control command is determined based on the type of obstacle, the distance to the obstacle, and the critical distance. Send the anti-pinch control command to the electric tailgate actuator. The anti-pinch control command is determined based on the obstacle type, the distance to the obstacle, and the critical distance, including: For dynamic flexible obstacles, when the distance to the obstacle is less than the critical distance, the anti-pinch control command is determined to control the tailgate motor to stop and then control the motor to rotate in the opposite direction by a predetermined angle. For both dynamic and static hard obstacles, when the distance to the obstacle is less than the critical distance, the anti-pinch control command is determined to stop the tailgate motor. For non-dangerous targets, no anti-pinch control commands are generated.

2. The method according to claim 1, characterized in that, The tailgate motion data includes the tailgate's velocity and acceleration, and the obstacle motion data includes the obstacle's moving speed. Determining the critical distance based on obstacle motion data, tailgate motion data, and predefined braking response time includes: A critical distance is determined based on the tailgate speed, tailgate acceleration, obstacle movement speed, and braking response time, wherein the critical distance is positively correlated with the tailgate speed, tailgate acceleration, obstacle movement speed, and braking response time.

3. The method according to any one of claims 1-2, characterized in that, The preprocessed data is used to identify and classify obstacles using an obstacle recognition and classification model to determine the types of obstacles, including: Through the feature extraction layer in the obstacle recognition and classification model, features of obstacles at multiple different scales are extracted from the preprocessed data to generate feature maps at multiple different scales. The obstacle recognition and classification model uses a classification prediction head to calculate the category probability distribution of the predicted anchor boxes on the feature map, and determines the obstacle type based on the category probability distribution.

4. The method according to any one of claims 1-2, characterized in that, The obstacle recognition and classification model is obtained through the following steps: Construct an anti-pinch scenario dataset for electric tailgates, wherein the samples in the anti-pinch scenario dataset include radar feature maps under various anti-pinch scenarios and the true categories of the labeled samples; The electric tailgate anti-pinch scenario dataset is input into the obstacle recognition and classification model, and the predicted category is output by the obstacle recognition and classification model; A loss function is generated based on the difference between the predicted category and the true category, as well as the category weight and / or feature scale weight, and the obstacle recognition classification model is trained using the loss function.

5. An anti-pinch control device for an electric tailgate, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the electric tailgate anti-pinch control method as described in any one of claims 1 to 4.

6. An anti-pinch control system for an electric tailgate, characterized in that, The system includes an obstacle sensing module, an electric tailgate actuator, and an electric tailgate anti-pinch control device according to claim 5, wherein... The obstacle sensing module includes a millimeter-wave radar sensor, which is installed at a position related to the movement of the electric tailgate of the vehicle. It is used to sense obstacles in the moving environment of the electric tailgate and generate raw radar point cloud data associated with the obstacles. The electric tailgate actuator includes a tailgate motor and a tailgate motion sensor, wherein the tailgate motion sensor is used to sense tailgate motion data, and the tailgate motor is used to execute anti-pinch control commands.

7. The system according to claim 6, characterized in that, The obstacle sensing module includes at least three millimeter-wave radar sensors, which are respectively arranged on the middle of the upper edge of the inner side of the electric tailgate and on both sides.

8. A vehicle, characterized in that, The vehicle includes an electric tailgate anti-pinch control system as described in claim 6 or 7.

Images