In-vehicle person detection method and device, vehicle and computer readable storage medium

Abstract

This invention proposes a method, device, vehicle, and computer-readable storage medium for detecting occupants inside a vehicle. By installing pressure sensors on the seats inside the vehicle, the system can cover the possible locations where occupants may be present, avoiding blind spots in detection. Furthermore, by arranging the pressure sensors in an array, the target detection model comprehensively analyzes the data detected by the pressure sensors in different distributions to detect occupants. This allows the system to perceive changes in occupant body shape and posture, accurately distinguishing between occupants and heavy objects. Based on current environmental data, the system clarifies the characteristics of the detection environment and determines the target judgment threshold by combining the influence of the environment on pressure detection, thereby improving the accuracy of occupant detection.

Description

Technical Field

[0001] This invention relates to the field of vehicle safety, and more particularly to a method, apparatus, vehicle, and computer-readable storage medium for detecting occupants inside a vehicle. Background Technology

[0002] If a child is left in a car, the child will be in danger. Therefore, in order to implement an alarm in such a scenario, existing technologies include child detection after parking. However, existing detection methods usually use cameras, radar, or seat belt tension-based detection. Camera and radar solutions have blind spots, while seat belt tension-based solutions have difficulty distinguishing between children and heavy objects. Therefore, the accuracy of existing technologies in detecting people inside the car is low. Summary of the Invention

[0003] The main objective of this invention is to provide a method, device, vehicle, and computer-readable storage medium for detecting occupants inside a vehicle, aiming to solve the problem of low accuracy in occupant detection in the prior art.

[0004] To achieve the above objectives, the present invention provides a method for detecting occupants inside a vehicle, the method comprising: Acquire pressure distribution data detected by a seat pressure sensor array, wherein the seat pressure sensor array includes multiple pressure sensors disposed on the seat; The probability of the presence of personnel is determined based on the pressure distribution data. Determine the target determination threshold based on the vehicle's current environmental data; When the probability of the presence of a person is greater than or equal to the target determination threshold, it is determined that there is a person in the vehicle.

[0005] Optionally, determining the probability of a person's presence based on the pressure distribution data includes: Obtain the seat status data of the seat; The pressure distribution data is compensated by the seat status data to obtain compensated pressure distribution data; The probability of the presence of the person is determined based on the compensated pressure distribution data.

[0006] Optionally, the step of compensating the pressure distribution data using the seat state data to obtain compensated pressure distribution data includes: Obtain the seat material and seat temperature from the seat status data; Match the correction value corresponding to the seat material and the seat temperature; The pressure distribution data is obtained by compensating the pressure distribution data with the correction value.

[0007] Optionally, determining the probability of a person's presence based on the pressure distribution data includes: Obtain the amount of allocable resources for the vehicle; Among the optional models, a target detection model corresponding to the amount of allocable resources is determined, wherein the accuracy of the target detection model is positively correlated with the amount of allocable resources; The pressure distribution data is input into the target detection model to obtain the probability of the presence of the person.

[0008] Optionally, determining the target detection model corresponding to the allocable resource quantity in the optional models includes: Obtain the processor load margin and remaining power in the allocable resource quantity; In the optional models, a target detection model is determined that corresponds to the processor load margin and the remaining power, wherein the accuracy of the target detection model is positively correlated with the processor load margin and the accuracy of the target detection model is positively correlated with the remaining power.

[0009] Optionally, the probability of the presence of personnel is obtained by inputting the pressure distribution data into the target detection model, and the method further includes: The results obtained from historical detection data identified high-risk samples indicating the presence of people inside the vehicle. Historical quantization output values ​​are obtained from the high-risk samples to obtain the historical quantization interval; The historical quantization interval is used as the focusing interval of the target detection model output value, and the corresponding quantization scaling factor is calculated based on the focusing interval, wherein the length of the focusing interval is positively correlated with the quantization scaling factor; The target detection model is updated according to the quantization scaling factor; The step of determining the probability of a person's presence based on the pressure distribution data includes: The pressure distribution data is input into the target detection model, so that the target detection model performs model quantization based on the quantization scaling factor and outputs the probability of the presence of the person corresponding to the pressure distribution data.

[0010] Optionally, the probability of the presence of personnel is obtained by inputting the pressure distribution data into the target detection model, and the determination of the target determination threshold based on the vehicle's current environmental data includes: Determine the adjustable threshold parameter corresponding to the current environmental data; Obtain the uncertainty of the target detection model output for the pressure distribution data; An initial threshold is obtained, and the target determination threshold is obtained by compensating the initial threshold with the adjustable threshold parameter, wherein the degree of compensation of the adjustable threshold parameter is positively correlated with the uncertainty.

[0011] To achieve the above objectives, the present invention also provides a vehicle occupant detection device, the vehicle occupant detection device comprising: The first acquisition module is used to acquire pressure distribution data detected by the seat pressure sensor array, wherein the seat pressure sensor array includes multiple pressure sensors disposed on the seat; The first determining module is used to determine the probability of the presence of personnel based on the pressure distribution data; The second determining module is used to determine the target determination threshold based on the vehicle's current environmental data; The third determining module is used to determine that there are people in the vehicle when the probability of the presence of the person is greater than or equal to the target determination threshold.

[0012] To achieve the above objectives, the present invention also provides a vehicle, the vehicle including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the vehicle occupant detection method as described above.

[0013] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the vehicle occupant detection method as described above.

[0014] This invention proposes a method, device, vehicle, and computer-readable storage medium for detecting occupants in a vehicle. The method involves acquiring pressure distribution data detected by a seat pressure sensor array, wherein the seat pressure sensor array includes multiple pressure sensors mounted on the seat. The method determines the probability of occupant presence based on the pressure distribution data; determines a target determination threshold based on the vehicle's current environmental data; and determines the presence of an occupant in the vehicle when the probability of occupant presence is greater than or equal to the target determination threshold. By installing pressure sensors on the vehicle seats, the method covers possible locations of occupants within the vehicle, avoiding detection blind spots. Furthermore, by arranging the pressure sensors in an array, the target detection model comprehensively analyzes data from pressure sensors with different distributions to detect occupants. This allows the model to perceive changes in occupant body shape and posture, accurately distinguishing between occupants and heavy objects. Based on current environmental data, the method clarifies the characteristics of the detection environment and, by combining the environmental influence on pressure detection, determines the target determination threshold, thereby improving the accuracy of occupant detection. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the vehicle occupant detection method of the present invention; Figure 2 This is a detailed flowchart of the vehicle occupant detection method of the present invention; Figure 3 This is a logical schematic diagram of the vehicle occupant detection method of the present invention; Figure 4 This is a schematic diagram of the modular structure of the vehicle of the present invention. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0019] This invention provides a method for detecting occupants inside a vehicle, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the vehicle occupant detection method of the present invention, the method comprising the following steps: Step S10: Obtain pressure distribution data detected by the seat pressure sensor array, wherein the seat pressure sensor array includes multiple pressure sensors disposed on the seat; Seats are the seats installed inside a vehicle, such as the driver's seat, the front passenger seat, and the rear seats.

[0020] The seat pressure sensor array is obtained by distributing multiple pressure sensors. The distribution of each pressure sensor in the seat pressure sensor array can be set according to actual needs, such as using a fixed horizontal interval and a fixed vertical interval as the distance between adjacent pressure sensors, to obtain a seat pressure sensor array in which the pressure sensors are arranged horizontally and vertically.

[0021] The specific type of pressure sensor can be set based on actual needs, such as selecting a pressure sensor with a sampling range of 0~100kPa and a resolution of 250P.

[0022] The timing of pressure distribution data acquisition can be set based on specific needs, such as continuous acquisition or data collection within a preset detection period after the vehicle is turned off. In scenarios where children or other individuals are left inside the vehicle after it is turned off, there is a high risk; therefore, the timing of detection can be set specifically for such scenarios. The specific length of the preset detection time can be set based on actual needs, such as 30 minutes after the vehicle is turned off. Within this preset detection time, pressure distribution data can be continuously collected or collected at intervals, such as acquiring pressure distribution data at preset intervals. The specific value of the preset interval can be set based on actual needs, such as 2 seconds. Taking the values ​​here as an example, pressure distribution data is collected every 2 seconds within 30 minutes after the vehicle is turned off, resulting in a total of 900 pressure distribution data collections within 30 minutes. After each pressure distribution data collection, subsequent detection steps can be performed based on the currently collected pressure distribution data. At the same time, it can also be combined with the pressure distribution data collected in the preset detection time to comprehensively perform subsequent detection steps. For example, if pressure distribution data is collected at the 4th second of the preset detection time, subsequent detection steps can be performed based on this pressure distribution data. At this time, the historical pressure distribution data detected at the 2nd second is also included. Therefore, subsequent detection steps can be performed by combining the pressure distribution data of the 2nd second and the 4th second. For example, the pressure distribution data of the 2nd second and the 4th second can both be input into the target detection model.

[0023] Pressure distribution data can be transmitted in the form of CAN messages.

[0024] To improve the accuracy of pressure distribution data, noise reduction can be performed on the pressure distribution data; the specific noise reduction method can be set according to actual needs.

[0025] Each pressure sensor in the seat pressure sensor array collects pressure values ​​at its location. The pressure values ​​collected by each pressure sensor are combined with their corresponding locations to obtain the pressure distribution data detected by the seat pressure sensor array. For example, if the seat pressure sensor array contains 1024 pressure sensors in 32 columns and 32 rows, the pressure distribution data contains pressure values ​​collected by 1024 pressure sensors, and these pressure values ​​are bound to the location of the corresponding pressure sensor, thus characterizing the pressure distribution at the location of the seat pressure sensor array. The above-mentioned number and distribution settings of pressure sensors are only for illustrative purposes. In practical applications, the parameters of the seat pressure sensor array can be set according to actual needs.

[0026] Pressure distribution data P raw It can be represented as:

[0027] Where i indicates the row in which the pressure sensor is located in the seat pressure sensor array, i∈[1,32]; j indicates the column in which the pressure sensor is located in the seat pressure sensor array, j∈[1,32].

[0028] The seat pressure sensor array is installed inside the seat, and the specific location can be set according to actual needs, such as setting the seat pressure sensor array inside the seat foam layer.

[0029] In this embodiment, by setting up a seat pressure sensor array, the pressure at different locations within the area covered by the seat pressure sensor array can be detected. This allows for the comprehensive assessment of the pressure distribution across the area to provide an overall picture of the seat pressure. Compared to single-point pressure detection, this provides more information and improves the accuracy of subsequent detection.

[0030] In practical applications, data collection can be further combined with radar, acoustics, cameras, and other methods for personnel detection. Radar methods, such as embedding miniature radar modules in the roof lining or seat backs, analyze the micro-Doppler effect of radar echoes to directly detect minute vibrations caused by chest rise and fall and heartbeat, enabling direct detection of vital signs with minimal interference from still objects and extremely high accuracy. However, these methods are more expensive, have more complex algorithms, and may raise privacy concerns. Acoustic methods, such as using ultrasonic sensor arrays to scan the contours of the seat surface to determine the presence of objects the size of a child, intermittently activate in-car microphones during sleep to detect audio features such as crying and breathing sounds. These methods use low-cost sensors and provide direct evidence through audio detection. However, they are susceptible to environmental noise, and ultrasound is sensitive to coverings such as thick blankets. Furthermore, continuous audio monitoring raises serious privacy concerns.

[0031] Step S20: Determine the probability of the presence of personnel based on the pressure distribution data; The probability of the presence of personnel is the probability that there may be personnel inside the vehicle, determined based on pressure distribution data. The specific method for determining the probability of the presence of personnel can be set according to actual needs, such as by comparing the pressure distribution data with the preset pressure distribution data when personnel are present, or by setting a detection model to determine the probability of the presence of personnel.

[0032] Step S30: Determine the target determination threshold based on the vehicle's current environmental data; Step S40: When the probability of the presence of a person is greater than or equal to the target determination threshold, it is determined that there is a person in the vehicle.

[0033] The target detection threshold is used to determine whether a person exists. If the probability of a person's presence reaches the target detection threshold, a detection result indicating the presence of a person is output; if the probability of a person's presence is less than the target detection threshold, a detection result indicating the absence of a person is output. A person can specifically include adults, children, and the elderly; this embodiment and subsequent embodiments use a child as an example for illustration. The target determination threshold is a threshold determined based on the current environmental data.

[0034] The correspondence between the target determination threshold and the current environmental data can be set based on the impact of specific data on personnel determination; for example, the specific type of current environmental data can be set based on actual needs, such as vehicle cabin temperature, current time, seat heating status, etc.

[0035] Higher vehicle cabin temperatures negatively impact the detection capabilities of pressure sensors, leading to reduced perception of occupants. Therefore, a lower target threshold needs to be set to improve detection accuracy; conversely, a higher target threshold needs to be set to reduce detection sensitivity.

[0036] The current time indicates the location within a day. It is understandable that during the day, due to the large number of people getting on and off vehicles, there are more issues with people being left behind. Therefore, daytime is a high-risk period for people being left behind. Thus, a lower target judgment threshold can be set to avoid missed detections and improve detection accuracy. Nighttime, on the other hand, is a low-risk period, and the threshold can be increased to avoid false alarms. In other words, the current time is negatively correlated with the target judgment threshold.

[0037] The seat heating status indicator shows the heating status of the seat. As explained above, the seat pressure sensor array is located inside the seat. Therefore, seat heating will introduce a lot of noise, which will increase the error of the pressure distribution data. Therefore, when the heating level indicated by the seat heating status is higher, a higher target judgment threshold can be set to suppress false alarms. That is, the seat heating level is positively correlated with the target judgment threshold.

[0038] The target judgment threshold is obtained by taking into account the impact of various current environmental data on the judgment threshold.

[0039] Once it is determined whether there are people inside the vehicle, the corresponding actions can be performed based on the result.

[0040] No additional operations will be performed if there are no people inside the vehicle.

[0041] When there are people inside the vehicle, both in-vehicle and remote alarms can be activated. The specific alarm method can be set according to actual needs. For example, the in-vehicle alarm can be set to sound an alarm via the in-vehicle buzzer and flash the vehicle's hazard lights. The remote alarm can send alarm information to the vehicle owner, the vehicle manufacturer, or relevant departments via mobile network. The specific content of the alarm information can be set according to actual needs, such as VIN (Vehicle Identification Number), time, and location.

[0042] The alarm method can also be classified based on the detection results. When the detection result indicates that a person has been detected and the probability of the person's presence is less than the preset alarm probability, a warning message is pushed. When the detection result indicates that a person has been detected and the probability of the person's presence is greater than or equal to the preset alarm probability, an alarm message is pushed. The value of the preset alarm probability can be set according to actual needs.

[0043] This embodiment uses pressure sensors installed on the seats inside the vehicle to cover all possible locations where people may be inside the vehicle, avoiding blind spots in detection. Furthermore, by arranging the pressure sensors in an array, the target detection model uses data from these differently distributed pressure sensors to comprehensively detect people. This allows it to perceive changes in body shape and posture, accurately distinguishing people from heavy objects. Based on current environmental data, it clarifies the characteristics of the detection environment and determines the target judgment threshold by combining the environmental impact on pressure detection. This meets the detection needs of different scenarios, improves the accuracy of detection in various scenarios, and ultimately enhances the accuracy of detecting people inside the vehicle.

[0044] Further details will follow. Figure 2 In the second embodiment of the vehicle occupant detection method of the present invention based on the first embodiment, step S20 includes the following steps: Step S21: Obtain the seat status data of the seat; Step S22: Compensate the pressure distribution data using the seat state data to obtain compensated pressure distribution data; Step S23: Determine the probability of the presence of the person based on the compensated pressure distribution data.

[0045] Seat status data reflects the seat's status in a real-time environment; the specific type of seat status data can be set based on the actual impact of the seat on pressure detection, such as seat material and temperature.

[0046] It is understandable that the seat pressure sensor array is located inside the seat. Therefore, the state of the seat will affect the detection of the pressure sensors, resulting in deviations in the pressure distribution data collected by the seat pressure sensor array. Ideally, if there are no people or other objects on the seat, the pressure value detected by each pressure sensor in the seat pressure sensor array should be 0, that is, the baseline of the pressure sensor detection is 0. However, due to the influence of factors such as the material and temperature of the seat, the pressure sensor may detect a non-zero, stable pressure value when there are no people, that is, the baseline of the pressure sensor is shifted. This causes the pressure value to also shift when there are people, resulting in deviations in pressure detection.

[0047] Therefore, in this embodiment, seat status data is acquired and used to compensate for pressure distribution data, thereby eliminating the interference of the seat on the pressure distribution data and obtaining compensated pressure data after interference elimination; then, personnel detection is performed using the compensated pressure data to obtain accurate personnel detection results.

[0048] Further, step S22 includes the following steps: Step S221: Obtain the seat material and seat temperature from the seat status data; Step S222: Match the correction value corresponding to the seat material and the seat temperature; Step S223: The pressure distribution data is compensated using the correction value to obtain the compensated pressure distribution data.

[0049] Seat material refers to the type of material used to manufacture the seat. Seat material can be set at the time of vehicle manufacturing based on the actual installation of the seat, or it can be updated after the seat material is changed.

[0050] The seat temperature is the real-time temperature of the seat; the seat temperature can be detected by setting a temperature detection device; for example, by setting a temperature sensor inside the seat or in the space where the seat is located.

[0051] It's understandable that seats made of different materials possess different characteristics, such as weight, elasticity, and static pressure. These different characteristics cause deformation under specific conditions, and the resulting compression or voids can affect the pressure sensor's readings. For example, leather seats expand when heated, and the degree of expansion varies at different temperatures. This leads to different readings from the pressure sensors installed within the seat under the same load-bearing conditions; for instance, the pressure sensor readings are generally higher at high temperatures and lower at low temperatures. Similarly, while foam seats also expand when heated, the degree and nature of this expansion differ from leather, thus affecting the pressure sensor readings differently.

[0052] The calibration value is a calibration parameter that corrects the pressure sensor's detection data to a value unaffected by the seat, based on the specific seat material and temperature.

[0053] The specific method for determining the correction value can be set based on actual needs, such as pre-setting a correspondence table between seat material, seat temperature, and correction value. Specifically, experiments can be conducted without personnel present, obtaining the pressure sensor's detection values ​​under different seat materials and temperatures. Since the ideal requirement is for the pressure sensor's detection value to be 0, the negative value of the pressure sensor at this time can be used as the correction value for the specific seat material and temperature, i.e., correction value + detection value = 0. By simulating multiple seat materials and temperatures, multiple sets of correspondences can be obtained. Then, in practical applications, the corresponding correction value can be obtained by matching the seat material and temperature in the correspondence table.

[0054] In addition to building a relational table, the model can also be trained by using seat material, seat temperature and corresponding correction values, and then the correction values ​​corresponding to seat material and seat temperature can be determined by the trained model; the model settings and training can be set according to actual needs.

[0055] After obtaining the correction value, the pressure distribution data can be corrected using the correction value; for example, for each pressure value in the pressure distribution data, the corresponding correction value is subtracted from the pressure value to obtain the compensation pressure value corresponding to the pressure value; and the compensation pressure data is obtained by combining all the compensation pressure values.

[0056] It should be noted that the seat material affects the pressure sensors at different locations differently. For example, for leather seats, due to the high tension of leather, the leather at the edges of the seat is relatively fixed, resulting in greater pressure from deformation, while the leather in the middle of the seat has more space, resulting in less pressure from deformation. Other materials will also have different effects on the pressure sensors at different locations due to their specific characteristics. Therefore, when determining the compensation pressure data, a corresponding relationship table can be set for each pressure sensor in the seat pressure sensor array based on the actual situation of its location to determine the calibration value of the pressure sensor at each location, thereby obtaining the compensation pressure value for each pressure sensor, and finally obtaining the compensation pressure data.

[0057] In other embodiments, additional parameters of the environment in which the seat is located can be incorporated to obtain compensation pressure data, such as seat humidity.

[0058] Furthermore, in the third embodiment of the vehicle occupant detection method of the present invention based on the first embodiment, step S20 includes the following steps: Step S24: Obtain the amount of allocable resources for the vehicle; Step S25: Determine the target detection model corresponding to the allocable resource quantity from the optional models, wherein the accuracy of the target detection model is positively correlated with the allocable resource quantity; Step S26: Input the pressure distribution data into the target detection model to obtain the probability of the presence of the person.

[0059] The object detection model is used to detect whether there are people inside a vehicle. The specific type of object detection model can be set according to actual needs, such as CNN (Convolutional Neural Network) model or Transformer-CNN hybrid model. It should be noted that the settings, training, and updating of the object detection model can be set according to actual needs.

[0060] After obtaining the pressure distribution data, the pressure distribution data is input into the target detection model so that the target detection model can detect whether there are people in the vehicle based on the pressure distribution at different locations, so as to obtain the probability of the presence of people; the probability of the presence of people specifically indicates whether there are people in the vehicle or whether there are no people in the vehicle.

[0061] The allocable resources are the remaining resources in the vehicle used to support the operation of the target detection model; the specific resource types of the allocable resources can be set based on actual needs, such as the processor's load margin and the vehicle's remaining battery power.

[0062] It is understandable that object detection models require vehicle resources when performing detection tasks; however, when the vehicle has limited available resources, enabling a high-precision object detection model can affect other vehicle functions or increase the rate of resource consumption, thus impacting the vehicle's range; if a low-precision object detection model is used, the detection accuracy will be lower.

[0063] Therefore, in order to balance vehicle resource utilization and target detection accuracy, this embodiment sets up optional models with different levels of precision.

[0064] The selectable models are the detection models that can be used in the vehicle. There are multiple selectable models, each with a different level of precision. The specific precision can be set according to actual needs, such as INT4, INT8, and FP16 from lowest to highest precision. The higher the precision of the selectable model, the higher the detection accuracy, but the more resources it consumes in the vehicle. Conversely, the lower the precision of the selectable model, the less resources it consumes, but the lower the detection accuracy. Therefore, in this embodiment, the amount of allocable resources for the vehicle is first obtained, and a suitable model is selected from the selectable models based on the amount of allocable resources. Specifically, the precision of the selected target detection model is positively correlated with the amount of allocable resources. That is, when the amount of allocable resources is less, a target detection model with lower precision is selected to avoid affecting other functions and range of the vehicle; when the amount of allocable resources is more, a target detection model with higher precision is selected to improve the accuracy of target detection.

[0065] Further, step S25 includes the following steps: Step S251: Obtain the processor load margin and remaining power in the allocable resources; Step S252: Determine the target detection model from the optional models that corresponds to the processor load margin and the remaining power, wherein the accuracy of the target detection model is positively correlated with the processor load margin and the accuracy of the target detection model is positively correlated with the remaining power.

[0066] In this embodiment, processor load margin and remaining power are used to quantify the amount of allocable resources.

[0067] The processor load margin is the remaining available load of the processor, which can be calculated by subtracting the current processor load from 1. It can be understood that the more processor load margin there is, the more tasks it can handle, and therefore, a more accurate target detection model can be used. Therefore, in this embodiment, the accuracy of the target detection model is set to be positively correlated with the processor load margin.

[0068] The remaining power is the vehicle's available power. It can be understood that the more available power, the longer the vehicle's driving time. The higher the power consumption of the target detection model, the lower its impact on the vehicle's driving time. Therefore, a more accurate target detection model can be used. Therefore, in this embodiment, the accuracy of the target detection model is set to be positively correlated with the remaining power.

[0069] To facilitate the determination of target detection models in practical applications, the correspondence between processor load margin, remaining power, and target detection models can be pre-defined; for example:

[0070] In scenarios with high load margin and high power consumption, the FP16 model with the highest accuracy is used as the target detection model, thereby obtaining the highest accuracy target detection model and ensuring the accuracy of target detection. In scenarios with high load margin and medium power consumption, the INT8 model with moderate accuracy is adopted as the target detection model, thereby obtaining a target detection model that balances accuracy and performance. In scenarios with high load margin and low battery, the INT4 model with the lowest accuracy is used as the target detection model, thereby obtaining the most power-efficient target detection model and ensuring vehicle range. In scenarios with medium load margin and high power consumption, the INT8 model with moderate accuracy is adopted as the target detection model, so as to obtain a target detection model that ensures response speed while balancing accuracy and performance. In scenarios with medium load margin and medium power consumption, the INT8 model with moderate accuracy is adopted as the target detection model, so as to obtain a stable target detection model while balancing accuracy and performance. In scenarios with medium load margin and low battery, the INT4 model with the lowest accuracy is used as the target detection model, which can reduce processor pressure, save power, and ensure vehicle range. In scenarios with low load margin and high power consumption, the INT4 model with the lowest accuracy is used as the target detection model, thereby freeing up processor performance for critical tasks and ensuring the implementation of other functions. In scenarios with low load margin and medium power consumption, the INT4 model with the lowest accuracy is used as the target detection model to avoid system lag. In scenarios with low load margin and low power consumption, the INT4 model with the lowest accuracy is used as the target detection model to avoid affecting other functions while achieving optimal power saving.

[0071] The load margin segments, power segments, and specific value settings in the table above are for illustrative purposes only. In actual applications, these settings can be configured based on actual needs.

[0072] Furthermore, in the fifth embodiment of the vehicle occupant detection method of the present invention based on the first embodiment, the method further includes the step of: Step S50: Obtain high-risk samples of people inside the vehicle from historical detection data; Step S60: Obtain historical quantization output values ​​from the high-risk samples to obtain the historical quantization range; Step S70: Use the historical quantization interval as the focusing interval of the target detection model output value, and calculate the corresponding quantization scaling factor based on the focusing interval, wherein the length of the focusing interval is positively correlated with the quantization scaling factor; Step S80: Update the target detection model according to the quantization scaling factor; Step S20 includes: Step S27: Input the pressure distribution data into the target detection model so that the target detection model performs model quantization based on the quantization scaling factor and outputs the probability of the presence of the person corresponding to the pressure distribution data.

[0073] Historical detection data refers to relevant data generated during actual personnel detection; such as pressure distribution data, personnel detection results, and internal parameters of the target detection model generated each time personnel detection is conducted, all of which can be included in the historical detection data corresponding to that detection. The internal parameters of the target detection model can include the probability of personnel presence, quantization scaling factor, judgment threshold, etc.

[0074] High-risk samples are sample data from historical detection data that show the presence of people inside vehicles; a sample data is a collection of relevant data generated in a single person detection, such as pressure distribution data, person detection results, and internal parameters of the target detection model.

[0075] The quantized output value is the probability of the presence of people as output by the target detection model.

[0076] The historical quantization output value is the probability of the presence of a person in a specific sample, as output by the target detection model.

[0077] The historical quantization interval is the smallest interval containing the historical quantization output values ​​corresponding to all high-risk samples. For example, if the historical quantization output values ​​include 0.6, 0.7, 0.75, 0.85, and 0.9, then the historical quantization interval is [0.6, 0.9], with the minimum and maximum values ​​among them as the interval endpoints. In practical applications, to avoid the historical quantization interval being too small due to a small number of samples, an interval margin can be set to extend both ends of the historical quantization interval by the interval margin. For example, if the interval margin is set to 0.5, then the historical quantization interval obtained based on the above historical quantization output values ​​is [0.55, 0.95].

[0078] Model quantization refers to converting parameters such as weights and activation values, which are originally stored and calculated in high-precision floating-point numbers (such as FP32 and FP16), into low-bit-width integers (such as INT8 and INT4) to reduce model size, lower computational overhead, and improve inference speed. Model quantization specifically includes quantization and dequantization. The quantization process converts floating-point numbers into low-precision integers based on a quantization scaling factor.

[0079] Where, x q x is the quantized integer value; f `float` is the number before conversion; `scale` is the quantization scaling factor; `zero_point` is the zero-point offset, which is an integer used to align the floating-point zero with the quantized integer zero; `round(·)` is the rounding function.

[0080] The dequantization process converts low-precision integers into floating-point numbers based on a quantization scaling factor.

[0081] Where, x dequant This is the floating-point number obtained after dequantization.

[0082] Understandably, because quantized integers inherently have lower precision, floating-point numbers will introduce errors after quantization and dequantization. Specifically, the quantization error is:

[0083] However, due to the limited number of values ​​for low-precision integers, such as INT8 having 255 values, when the floating-point number to be quantized is the probability of a person's existence, the probability of a person's existence is mapped to the interval [0, 100%], and the precision of the probability of a person's existence is 1 / 255. Here, we take the probability of a person's existence as an example; the other parameters involved in the model quantization are similar.

[0084] However, the probability of personnel presence reflects a certain degree of concentration. For example, when the focus interval is [0, 100%], the quantized integers are uniformly distributed within the [0, 100%] interval. However, for personnel detection, the quantized output values ​​that affect the final judgment are distributed within a smaller decision-sensitive interval, specifically, within the interval near the judgment threshold. Therefore, the accuracy of the quantized integers within this focus interval is low, leading to misjudgments. For example, for the same pressure distribution data, the probability of personnel presence obtained after quantization is 0.65, while the high-precision FP32 model without quantization outputs a probability of personnel presence of 0.78. The corresponding quantization error is 0.78 - 0.65 = 0.13. If the judgment threshold is between 0.65 and 0.78, it will lead to two different detection results, which is very easy to cause misjudgments.

[0085] Therefore, in this embodiment, the historical quantization interval is used as the focus interval to determine the quantization scaling factor, so that a smaller quantization scaling factor can be generated based on a narrower focus interval, thereby improving the accuracy of the quantized integer.

[0086] The focus interval is the mapping interval for calculating the quantization scaling factor. Taking INT8 as an example, the number of values ​​is 255; the focus interval is [a, b], then the quantization scaling factor scale has:

[0087] After determining the quantization scaling factor based on the focus interval, model quantization can be performed based on the quantization scaling factor, thereby improving the quantization resolution of the focus interval. For example, if the focus interval determined based on the historical quantization interval is [0.4, 0.9], then the obtained quantization scaling factor is 1 / 510. At this time, within the focus interval, the interval between adjacent integers after quantization is 1 / 510, which is 1 / 255 compared to the [0, 1] focus interval, thus doubling the quantization resolution.

[0088] It should be noted that zero-point offset has a corresponding valid range based on model accuracy. For example, the valid range of zero-point offset in the INT8 model is [ [128, 127]; However, after updating the quantization scaling factor, the zero-point offset may exceed the legal range. Therefore, to avoid this problem, one can choose to retain FP16 precision for the final output layer in the model and only update the quantization scaling factor of INT8 / INT4 for the intermediate computation layers; symmetric quantization can also be used, such as setting the zero-point offset to 0 and clipping the output to [0, 0.9]. In this case, scale = 0.9 / 127, ensuring that all values ​​can be legally represented. However, this reduces the resolution of the focusing interval. Therefore, specific methods can be selected based on actual needs to prevent the zero-point offset from exceeding the legal range.

[0089] See Figure 3 To further improve the accuracy of quantization scaling factor updates, historical detection data can be uploaded to the cloud. A high-precision model in the cloud can then use this data to obtain a high-precision probability of human presence and a high-precision historical quantization interval. The quantization scaling factor is then calculated based on this interval and sent to the vehicle-side to update the target detection model's quantization scaling factor. Specifically, the updated quantization scaling factor and zero-point offset table can be sent to the vehicle-side via secure OTA (Over-The-Air) updates in a <1KB differential parameter package. The vehicle-side MCU hot-loads the new parameters during runtime without restarting the model, and the changes take effect immediately in the next inference cycle. Compared to the resources available on the vehicle-side, the cloud has the resources to deploy higher-precision models. Therefore, calculating the quantization scaling factor using a high-precision model based on the cloud improves its accuracy. The specific type and precision of the high-precision model can be set according to actual needs, such as using an FP32 precision model.

[0090] In other embodiments, the target personnel detection step can be set up in the cloud. That is, after the vehicle collects pressure distribution data, it uploads the pressure distribution data to the cloud for detection. The cloud then returns the personnel detection results to the vehicle. In this way, since the cloud model has no resource limitations, a more accurate model can be set up to improve the accuracy rate, and the hardware cost on the terminal side is greatly reduced. At the same time, model updates do not require OTA and can be completed directly in the cloud. However, this method is highly dependent on the network and may fail in underground parking garage scenarios with poor signal. Moreover, the data upload traffic is large, which raises privacy and cost issues. Alarm delay is affected by the network.

[0091] Furthermore, in the sixth embodiment of the vehicle occupant detection method of the present invention based on the first embodiment, the probability of occupant presence is obtained by inputting the pressure distribution data into the target detection model, and step S50 includes: Step S51: Determine the adjustable threshold parameter corresponding to the current environmental data; Step S52: Obtain the uncertainty of the target detection model output for the pressure distribution data; Step S53: Obtain an initial threshold, and compensate the initial threshold with the adjustable threshold parameter to obtain the target determination threshold, wherein the degree of compensation of the adjustable threshold parameter is positively correlated with the uncertainty.

[0092] The adjustable threshold parameter is a parameter that is adjusted based on environmental data during the determination of the target judgment threshold.

[0093] like:

[0094] Where K is an adjustable threshold parameter; k0 is an initial parameter, the specific value of which can be set according to actual needs, such as 1.2; f(E, t) is the current environmental data, where E is the environmental vector and t is the time vector; temperature is the normalized vehicle cabin temperature; α is the weight of the vehicle cabin temperature; time_of_day is the normalized current time, β is the weight of the current time; seat_heating_status is the normalized seat heating status; γ is the weight of the seat heating status; the specific values ​​of α, β, and γ can be set according to actual needs, such as α=-0.3 (sign indicates negative correlation); β=-0.8 (sign indicates negative correlation); γ=0.5.

[0095] Taking a summer afternoon with a vehicle cabin temperature of 35°C, seat heating on high, and the current time 2:00 PM as an example; first, normalize the parameters: For vehicle cabin temperature, normalization is performed using a temperature range of -10℃ to 60℃ as an example: temperature = (35 + 10) / 60 = 0.75 Normalize the current time: time_of_day = 14 / 24 ≈ 0.583 For seat heating settings, taking the three levels of low, medium, and high as an example, high setting is normalized to 1; low setting is normalized to 0; and medium setting is normalized to 0.5.

[0096] Adjustable threshold parameters include: K =1.2 + (-0.3)×0.75 + (-0.8)×0.583 + 0.5×1.0 = 1.009 Let's take a cold winter night as an example, with the vehicle cabin temperature at 0°C, seat heaters off, and the current time at 2:00 AM; first, normalize the parameters: For vehicle cabin temperature, normalization is performed using a temperature range of -10℃ to 60℃ as an example: temperature = (0 + 10) / 60 ≈ 0.167 Normalize the current time: time_of_day = 2 / 24 ≈ 0.083 Regarding the seat heating status, taking the three levels of low, medium, and high as an example, the low level is normalized to 0.

[0097] Adjustable threshold parameters include: K =1.2 + (-0.3)×0.167 + (-0.8)×0.083 + 0.5×0= 1.083 In the case of daytime high temperature and heating, K=1.009, the adjustable threshold parameter is low, which makes it easy to trigger alarms and prevent false alarms; In a cold, late-night scenario without heating, K=1.083, which is a relatively high adjustable threshold parameter and is less prone to false alarms.

[0098] The target determination threshold is determined by combining the initial threshold, adjustable threshold parameters, and uncertainty, and is as follows:

[0099] Where T is the target determination threshold; σ is the initial threshold, and the specific value can be set according to actual needs, such as 0.7; σ is the uncertainty, which is output by the model itself. For example, after each detection, the model will output the probability of the presence of people P and the uncertainty σ; the probability of the presence of people is the quantified probability of the target detection model's understanding of the presence of people based on its own knowledge representation of the pressure distribution data. The more obvious the presence of people reflected in the pressure distribution data, the greater the probability of the presence of people output by the target detection model; the uncertainty σ indicates the target detection model's confidence in the output probability of the presence of people. The lower the uncertainty, the more confident the target detection model is in the correctness of the output probability of the presence of people; the specific method of determining the uncertainty can be set according to actual needs, such as Monte Carlo Dropout or deep ensemble.

[0100] The greater the uncertainty, the lower the accuracy of the probability of the presence of people output by the target detection model, and the more objective data is needed for compensation. Therefore, the degree of compensation of the initial judgment threshold by the adjustable threshold parameter is positively correlated with the uncertainty. The greater the uncertainty, the greater the influence of the adjustable threshold parameter in the determination of the initial judgment threshold, and the more the initial judgment threshold can be adjusted based on the objective environment, thereby improving the detection accuracy.

[0101] To further improve the accuracy of determining the judgment threshold, false alarm samples and missed alarm samples from historical monitoring data can be obtained. Based on these samples, the adjustable threshold parameter can be adjusted. During the adjustment process, the adjustable threshold parameter with a false positive rate (FPR) of less than 0.1% and the smallest FPR can be determined. This adjustable threshold parameter is then applied to update the initial parameters in the adjustable threshold parameters. The specific adjustment method can be set according to actual needs, such as context-aware adjustment.

[0102] In this embodiment, the degree of compensation for the adjustable threshold parameter is positively correlated with the uncertainty, so that the target judgment threshold can be accurately determined.

[0103] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0105] This application also provides an in-vehicle occupant detection device for implementing the above-described in-vehicle occupant detection method, the in-vehicle occupant detection device comprising: The first acquisition module is used to acquire pressure distribution data detected by the seat pressure sensor array, wherein the seat pressure sensor array includes multiple pressure sensors disposed on the seat; The first determining module is used to determine the probability of the presence of personnel based on the pressure distribution data; The second determining module is used to determine the target determination threshold based on the vehicle's current environmental data; The third determining module is used to determine that there are people in the vehicle when the probability of the presence of the person is greater than or equal to the target determination threshold.

[0106] This in-vehicle occupant detection device uses pressure sensors installed on the seats to cover potential locations of children inside the vehicle, avoiding blind spots. Furthermore, by arranging the pressure sensors in an array, the target detection model uses data from the different distributions of pressure sensors to comprehensively detect children. This allows the device to perceive changes in the child's body shape and posture, accurately distinguishing children from heavy objects and improving the accuracy of occupant detection.

[0107] It should be noted that the first acquisition module in this embodiment can be used to execute step S10 in this application embodiment, the first determination module in this embodiment can be used to execute step S20 in this application embodiment, the second determination module in this embodiment can be used to execute step S30 in this application embodiment, and the third determination module in this embodiment can be used to execute step S40 in this application embodiment.

[0108] Furthermore, the first determining module includes: The first acquisition unit is used to acquire the seat status data of the seat; The first compensation unit is used to compensate the pressure distribution data using the seat state data to obtain compensated pressure distribution data. The first input unit is used to determine the probability of the presence of the person based on the compensated pressure distribution data.

[0109] Furthermore, the first compensation unit includes: The first acquisition subunit is used to acquire the seat material and seat temperature from the seat status data; The first matching subunit is used to match the correction value corresponding to the seat material and the seat temperature; The first compensation subunit is used to compensate the pressure distribution data using the correction value to obtain the compensated pressure distribution data.

[0110] Furthermore, the first determining module includes: The second acquisition unit is used to acquire the amount of allocable resources for the vehicle; The first determining unit is configured to determine, from the optional models, a target detection model corresponding to the allocable resource quantity, wherein the accuracy of the target detection model is positively correlated with the allocable resource quantity; The third acquisition unit is used to input the pressure distribution data into the target detection model to obtain the probability of the presence of the person.

[0111] Further, the first determining unit includes: The second acquisition subunit is used to acquire the processor load margin and remaining power in the allocable resource quantity; A first determining subunit is configured to determine, from the optional models, the target detection model corresponding to the processor load margin and the remaining power, wherein the accuracy of the target detection model is positively correlated with the processor load margin and the accuracy of the target detection model is positively correlated with the remaining power.

[0112] Furthermore, the probability of the presence of personnel is obtained by inputting the pressure distribution data into the target detection model, and the device further includes: The third acquisition module is used to obtain high-risk samples from historical detection data that indicate the presence of people inside the vehicle; The fourth acquisition module is used to acquire historical quantization output values ​​from the high-risk samples to obtain the historical quantization range; The first calculation module is used to use the historical quantization interval as the focusing interval of the target detection model output value, and to calculate the corresponding quantization scaling factor based on the focusing interval, wherein the length of the focusing interval is positively correlated with the quantization scaling factor. The first update module is used to update the target detection model according to the quantization scaling factor; The first determining module includes: The second input unit is used to input the pressure distribution data into the target detection model, so that the target detection model outputs the probability of the presence of the person corresponding to the pressure distribution data after model quantization based on the quantization scaling factor.

[0113] Furthermore, the probability of the presence of personnel is obtained by inputting the pressure distribution data into the target detection model, and the second determining module includes: The second determining unit is used to determine the adjustable threshold parameter corresponding to the current environmental data; The fifth acquisition unit is used to acquire the uncertainty of the target detection model output for the pressure distribution data; The sixth acquisition unit is used to acquire an initial threshold and compensate the initial threshold with the adjustable threshold parameter to obtain the target determination threshold, wherein the degree of compensation of the adjustable threshold parameter is positively correlated with the uncertainty.

[0114] Reference Figure 4 In terms of hardware structure, the vehicle may include components such as a communication module 10, a memory 20, and a processor 30. In the vehicle, the processor 30 is connected to both the memory 20 and the communication module 10. The memory 20 stores a computer program, which is executed by the processor 30. When the computer program is executed, it implements the steps of the above-described method embodiments.

[0115] The communication module 10 can connect to external communication devices via a network. The communication module 10 can receive requests from the external communication devices and can also send requests, instructions, and information to the external communication devices, which can be other vehicles, servers, or IoT devices, such as televisions, etc.

[0116] The memory 20 can be used to store software programs and various data. The memory 20 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as acquiring pressure distribution data detected by the seat pressure sensor array), etc.; the data storage area may include a database, and may store data or information created based on system usage. Furthermore, the memory 20 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.

[0117] The processor 30 is the control center of the vehicle. It connects to various parts of the vehicle via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 20, and by calling data stored in the memory 20, it performs various vehicle functions and processes data, thereby providing overall vehicle monitoring. The processor 30 may include one or more processing units; optionally, the processor 30 may integrate an application processor and a modem processor. The application processor primarily handles the operating system, user interface, and applications, while the modem processor primarily handles wireless communication. It is understood that the modem processor may not be integrated into the processor 30.

[0118] although Figure 4 Not shown, but the vehicle described above may also include a circuit control module for connecting to a power source to ensure the normal operation of other components. Those skilled in the art will understand that... Figure 4 The vehicle structure shown does not constitute a limitation on the vehicle and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0119] The present invention also proposes a computer-readable storage medium having a computer program stored thereon. The computer-readable storage medium may be... Figure 4 The memory 20 in the vehicle may be at least one of ROM (Read-Only Memory) / RAM (Random Access Memory), magnetic disk, optical disk, etc. The computer-readable storage medium includes a number of instructions to cause a terminal device with a processor (which may be a television, automobile, mobile phone, computer, server, terminal, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0120] In this invention, the terms "first," "second," "third," "fourth," and "fifth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0121] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0122] Although embodiments of the present invention have been shown and described above, the scope of protection of the present invention is not limited thereto. It is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, and substitutions to the above embodiments within the scope of the present invention, and such changes, modifications, and substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting occupants inside a vehicle, characterized in that, The method for detecting occupants inside the vehicle includes: Acquire pressure distribution data detected by a seat pressure sensor array, wherein the seat pressure sensor array includes multiple pressure sensors disposed on the seat; The probability of the presence of personnel is determined based on the pressure distribution data. Determine the target determination threshold based on the vehicle's current environmental data; When the probability of the presence of a person is greater than or equal to the target determination threshold, it is determined that there is a person in the vehicle.

2. The vehicle occupant detection method as described in claim 1, characterized in that, The step of determining the probability of a person's presence based on the pressure distribution data includes: Obtain the seat status data of the seat; The pressure distribution data is compensated by the seat status data to obtain compensated pressure distribution data; The probability of the presence of the person is determined based on the compensated pressure distribution data.

3. The vehicle occupant detection method as described in claim 2, characterized in that, The step of compensating the pressure distribution data using the seat state data to obtain the compensated pressure distribution data includes: Obtain the seat material and seat temperature from the seat status data; Match the correction value corresponding to the seat material and the seat temperature; The pressure distribution data is obtained by compensating the pressure distribution data with the correction value.

4. The vehicle occupant detection method as described in claim 1, characterized in that, The step of determining the probability of a person's presence based on the pressure distribution data includes: Obtain the amount of allocable resources for the vehicle; Among the optional models, a target detection model corresponding to the amount of allocable resources is determined, wherein the accuracy of the target detection model is positively correlated with the amount of allocable resources; The pressure distribution data is input into the target detection model to obtain the probability of the presence of the person.

5. The vehicle occupant detection method as described in claim 4, characterized in that, The step of determining the target detection model corresponding to the allocable resource quantity from the optional models includes: Obtain the processor load margin and remaining power in the allocable resource quantity; In the optional models, a target detection model is determined that corresponds to the processor load margin and the remaining power, wherein the accuracy of the target detection model is positively correlated with the processor load margin and the accuracy of the target detection model is positively correlated with the remaining power.

6. The vehicle occupant detection method as described in claim 1, characterized in that, The probability of the presence of personnel is obtained by inputting the pressure distribution data into the target detection model, and the method further includes: The results obtained from historical detection data identified high-risk samples indicating the presence of people inside the vehicle. Historical quantization output values ​​are obtained from the high-risk samples to obtain the historical quantization interval; The historical quantization interval is used as the focusing interval of the target detection model output value, and the corresponding quantization scaling factor is calculated based on the focusing interval, wherein the length of the focusing interval is positively correlated with the quantization scaling factor; The target detection model is updated according to the quantization scaling factor; The step of determining the probability of a person's presence based on the pressure distribution data includes: The pressure distribution data is input into the target detection model, so that the target detection model performs model quantization based on the quantization scaling factor and outputs the probability of the presence of the person corresponding to the pressure distribution data.

7. The method for detecting occupants inside a vehicle as described in claim 1, characterized in that, The probability of the presence of personnel is obtained by inputting the pressure distribution data into the target detection model. Determining the target determination threshold based on the vehicle's current environmental data includes: Determine the adjustable threshold parameter corresponding to the current environmental data; Obtain the uncertainty of the target detection model output for the pressure distribution data; An initial threshold is obtained, and the target determination threshold is obtained by compensating the initial threshold with the adjustable threshold parameter, wherein the degree of compensation of the adjustable threshold parameter is positively correlated with the uncertainty.

8. A vehicle occupant detection device, characterized in that, The vehicle occupant detection device includes: The first acquisition module is used to acquire pressure distribution data detected by the seat pressure sensor array, wherein the seat pressure sensor array includes multiple pressure sensors disposed on the seat; The first determining module is used to determine the probability of the presence of personnel based on the pressure distribution data; The second determining module is used to determine the target determination threshold based on the vehicle's current environmental data; The third determining module is used to determine that there are people in the vehicle when the probability of the presence of the person is greater than or equal to the target determination threshold.

9. A vehicle, characterized in that, The vehicle includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the occupant detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the vehicle occupant detection method as described in any one of claims 1 to 7.

Images