Temperature sensing coupling pedestrian flexibility test method for night AEB pedestrian test

Abstract

The invention relates to the technical field of road traffic networks and multi-vehicle cooperation, and provides a temperature-sensing coupling pedestrian flexibility testing method for night AEB pedestrian testing, which comprises the following steps: A, setting a testing environment for night AEB pedestrian testing, and deploying a flexible dummy in the testing environment; b, setting behavior logic of the flexible dummy meeting the AEB pedestrian test; c, the test vehicle obtains test data of the flexible dummy through the sensor assembly; d, constructing a temperature sensing compensation model based on the thermal radiation characteristics of the specific part, and carrying out real-time dynamic temperature compensation on the flexible dummy; and E, controlling the flexible dummy to move irregularly in the test environment, and calculating and recording a night AEB pedestrian test result in combination with test data when the AEB system triggers braking. The real performance of the AEB system at night can be accurately and objectively reflected.

Description

Technical Field

[0001] This invention relates to the field of vehicle braking technology, and more specifically to a temperature-sensing coupled pedestrian flexibility testing method for nighttime AEB pedestrian testing. Background Technology

[0002] With the rapid development of intelligent driving technology, the Automatic Emergency Braking (AEB) system is playing an increasingly crucial role in improving road traffic safety. Especially in nighttime driving scenarios, due to the significant reduction in visible light, the driver's visual perception range and judgment ability are severely limited. This places higher demands on the environmental perception capabilities of autonomous driving systems—they must be able to identify pedestrians, non-motorized vehicles, and other potential obstacles in a timely and accurate manner under low-light conditions and respond quickly.

[0003] One of the core functions of an Automatic Emergency Braking (AEB) system is to automatically trigger braking intervention when a potential collision risk occurs between a vehicle and a pedestrian ahead, in order to avoid a collision or significantly mitigate its consequences. However, current mainstream AEB systems still face significant challenges in nighttime pedestrian detection. For example, in the prior art, patent application CN202511239094.2 describes an AEB collision dummy adaptive control system and method based on vehicle-to-everything (V2X) technology. This system includes a collision dummy that switches between different body shapes based on dummy control commands and moves or stays on a test road segment based on its specific body shape. A test vehicle performs collision tests on the test road segment based on vehicle control commands. The test module first constructs a first test scenario, adjusts the collision dummy to the first body shape state, controls the collision dummy to move or stay on the test road segment, and uses the test vehicle to perform collision tests in the first test scenario. Then, it constructs an Nth test scenario, adjusts the collision dummy to the Nth body shape state, and uses the test vehicle to perform collision tests in the Nth test scenario. Currently, existing technologies primarily focus on adjusting the dummy's posture and shape to accurately identify the type of dummy. However, current technologies rely solely on accurate object identification. This is because existing solutions primarily depend on sensors such as cameras, millimeter-wave radar, and lidar for environmental perception. However, the performance of these sensors often degrades significantly under complex conditions such as low light, backlight, rain, fog, or strong glare. Visible light cameras are susceptible to insufficient light, lidar's sensitivity to low-reflectivity targets (such as pedestrians in dark clothing) decreases, and while millimeter-wave radar offers all-weather capability, it has limitations in target classification and precise positioning. Therefore, current technologies are often more focused on identifying the type of object.

[0004] To further improve the accuracy of pedestrian detection and enhance pedestrian recognition capabilities at night and in low visibility conditions, infrared thermal imaging (temperature sensing) technology has been introduced into recent research. This technology effectively distinguishes pedestrians or animals with body temperature characteristics from the background environment by detecting the thermal radiation characteristics emitted by objects, enhancing the system's robustness in dark, smoky, or foggy environments. However, a key technical bottleneck remains in AEB (Autonomous Emergency Braking) nighttime pedestrian testing based on infrared thermal imaging: the test dummies typically have built-in heating modules to simulate human surface temperature (approximately 37°C). However, in low-temperature environments (such as below -10°C), the dummy's surface temperature drops rapidly and unevenly due to strong convection and radiation, causing nonlinear attenuation or even distortion of the thermal radiation signal received by the infrared sensor. This dynamic temperature change severely affects the contrast and outline clarity of pedestrian targets in infrared images, thus interfering with the detection, tracking, and decision-making logic of the AEB system, ultimately leading to unreliable test results or system response delays. Therefore, current technologies primarily focus on identifying object types. Summary of the Invention

[0005] The present invention aims to provide a temperature-sensing coupled pedestrian flexibility testing method and system for nighttime AEB pedestrian testing, which can realize the dynamic thermal response of the flexible dummy under complex behavior, thereby more accurately and objectively reflecting the real performance of the AEB system at night.

[0006] The first scheme is a temperature-sensing coupled pedestrian flexibility testing method for nighttime AEB pedestrian testing, comprising: A: setting up a test environment for nighttime AEB pedestrian testing and deploying a flexible dummy within the test environment; B: setting behavioral logic for the flexible dummy under AEB pedestrian testing; C: deploying a test vehicle in the test environment, the test vehicle being equipped with sensor components to acquire test data of the flexible dummy in front of the test vehicle; wherein, the test data includes surface temperature distribution data and spatial position data of the flexible dummy; D: constructing a temperature-sensing compensation model based on the thermal radiation characteristics of specific parts, deploying the temperature-sensing compensation model in the flexible dummy, using the temperature-sensing compensation model to receive test data, generating temperature compensation strategies for characteristic parts of the flexible dummy, and performing real-time dynamic temperature compensation for the flexible dummy; E: controlling the flexible dummy to perform irregular movements in the test environment, and when the AEB system triggers braking, calculating and recording the nighttime AEB pedestrian test results based on the test data; wherein, the AEB pedestrian test results include the similarity of the test data of the flexible dummy, the consistency of the flexible dummy's movement trajectory, and the evaluation index results of the AEB pedestrian test.

[0007] Beneficial Effects: By establishing a complete process including test environment setup, dummy behavior setting, multi-source data acquisition, dynamic temperature compensation, and comprehensive result quantification, this system systematically solves the problems of dummy thermal radiation distortion, simplistic behavior simulation, and one-sided test results in nighttime AEB testing. The temperature compensation model enables real-time dynamic temperature control of specific parts of the flexible dummy, combined with irregular movements to simulate real pedestrian behavior. Furthermore, multi-dimensional indicators (data similarity, trajectory consistency, and core system performance) are used to achieve full-chain quantification, ensuring that the test results accurately and objectively reflect the real performance of the AEB system in complex nighttime scenarios, filling the gap in existing technologies regarding both thermal simulation and behavioral realism.

[0008] Preferably, the outermost layer of the flexible dummy is made of non-reflective material or is dressed in dark clothing; the composite thermal structure includes an inner thermal structure, a middle thermal structure, and an outer thermal structure in sequence; wherein, the inner thermal structure is a heating pipe wrapped with a thermally conductive silicone substrate, used to circulate thermally conductive oil and provide heat; the middle thermal structure is an aerogel insulation layer, used to block low-temperature interference from the outside of the flexible dummy; the outer thermal structure is an infrared characteristic simulation coating, which is made of human skin-like infrared material and used to provide human-like thermal radiation to the infrared thermal imager; the thermal conductivity of the thermally conductive silicone substrate is greater than 3 watts per meter per Kelvin.

[0009] Beneficial Effects: The design of a flexible, non-reflective / dark-colored outer layer and a three-layer composite thermal structure precisely solves the problem of thermal radiation distortion in low-temperature nighttime environments. The outer layer, a human-skin-like infrared coating, provides human-like thermal radiation characteristics, while the non-reflective / dark-colored clothing avoids light interference, improving the accuracy of infrared sensor recognition. The middle aerogel insulation layer effectively blocks external low temperatures, reducing heat loss. The inner layer is made of high thermal conductivity silicone (thermal conductivity > 3W / (m²)). K)) Wrap the heating pipes to ensure uniform heat transfer from the heat transfer oil, keeping the dummy's surface temperature stable within the human body temperature range of 36-38℃. Even at extreme low temperatures, it can maintain thermal stability, significantly improving the realism and reliability of the test scenario.

[0010] Preferably, the behavioral logic of the flexible dummy includes linear acceleration and deceleration, curvilinear motion, and random motion; wherein, linear acceleration and deceleration is used to simulate the scenario of pedestrians crossing the road; curvilinear motion and random motion are used to simulate unpredictable pedestrian behavior.

[0011] Beneficial Effects: By setting three behavioral logics—linear acceleration / deceleration, curvilinear motion, and random motion—the limitations of existing single-mode dummy movement are overcome. Linear acceleration / deceleration accurately simulates typical scenarios such as pedestrians crossing the road or suddenly appearing, while curvilinear and random motions reproduce unpredictable behaviors such as sudden turns and stops, comprehensively covering the movement characteristics of real pedestrians. The diversified behavioral logic increases the complexity and challenge of the test scenarios, fully exposing the shortcomings of the AEB system in detecting, tracking, and braking dynamic and sudden targets. This allows for a more comprehensive test of the system's emergency response capabilities and avoids test biases caused by adapting to fixed trajectories.

[0012] Preferably, the sensor assembly includes a thermal imager and a radar. The thermal imager is used to perceive the thermal radiation in front of the test vehicle in real time and convert the thermal radiation into temperature to obtain surface temperature distribution data of the flexible dummy in front of the vehicle. The surface temperature distribution data includes two-dimensional thermal radiation image data and a timestamp of image acquisition. The radar is used to perceive the position in front of the test vehicle in real time and obtain spatial position data of the flexible dummy in front of the vehicle. The spatial position data includes spatial point cloud data and a timestamp of point cloud data acquisition. The spatial point cloud data includes the three-dimensional position coordinates and velocity of the flexible dummy.

[0013] Beneficial effects: By combining thermal imagers and radar sensors, complementary acquisition of nighttime test data was achieved: the thermal imager accurately captured the surface temperature distribution and two-dimensional thermal radiation images of the dummy, solving the target identification problem caused by insufficient visible light at night; the radar acquired the dummy's three-dimensional position coordinates, velocity, and other spatial location data in real time, compensating for the limitations of the thermal imager in precise positioning. The two types of data, combined with timestamp recording, provided comprehensive and synchronous raw data support for subsequent time-series alignment, temperature compensation, and test result calculation, effectively avoiding the performance degradation problem of a single sensor under complex nighttime conditions, and improving the integrity and reliability of data acquisition.

[0014] Preferably, temporal alignment of surface temperature distribution data and spatial location data includes: extracting spatial point cloud data from the spatial location data as observation data, and performing temporal alignment of the spatial point cloud data using Kalman filtering. The temporal alignment process includes: constructing a state transition model and an observation model for a flexible dummy, and providing initial state data and initial covariance at an initial timestamp; for the state data at the current timestamp, performing state transition using a state transition model based on timestamp information to obtain state transition information, and sequentially executing a prediction stage and an update stage, wherein the prediction stage sequentially performs state prediction on the state transition information to obtain state transition information. In the covariance prediction stage, the observation model is used to convert the observation data into observation information. Combined with the state prediction results, covariance prediction results and observation information, Kalman gain, state update and covariance update are performed on the state transition information in sequence. The state update result is used as the alignment update result of the spatial point cloud data at the current timestamp. The state update result and covariance update result are used in the subsequent prediction stage. The alignment update result at each timestamp is extracted and concatenated with the timestamp to serve as the spatial location data after time-series alignment with the surface temperature distribution data. The surface temperature distribution data is assumed to be the time-series aligned data.

[0015] Beneficial Effects: By using Kalman filtering to achieve temporal alignment of sensor data, the asynchronous data problem caused by hardware response and transmission delays between thermal imagers and radar is specifically addressed. By constructing state transition and observation models, and combining prediction and update iteration processes, the timestamp errors of the two types of data are corrected to within milliseconds, ensuring accurate matching between the dummy's temperature distribution and spatial location data at the same moment. The temporally aligned dataset provides a spatiotemporally synchronized input foundation for the temperature sensing compensation model, avoiding misjudgments caused by data misalignment, improving the accuracy of test result calculations, and reducing the interference of temporal deviations on the performance evaluation of the AEB system.

[0016] Preferably, the state transition model in the Kalman filtering process is optimized, and the optimized state transition model is calculated as follows:

[0017] ; ; ; in, This represents the alignment update result of the spatial point cloud data at timestamp t-1. Representing spatial point cloud data The three-dimensional position coordinates in Representing spatial point cloud data The speed in the middle, This represents the state transition matrix corresponding to timestamp t. Represents the L2 norm. Indicates speed control parameters; This represents the noise parameters at timestamp t-1. This indicates that the mean is 0 and the covariance is... Multidimensional Gaussian distribution, Indicates the time interval between adjacent timestamps. This represents the variance of acceleration of the flexible pedestrians used in the nighttime AEB pedestrian test. This represents the state transition information at timestamp t.

[0018] Beneficial Effects: By optimizing the state transition model of the Kalman filter and incorporating key parameters such as velocity control parameters, noise suppression terms, and acceleration variance, the accuracy and adaptability of time-series alignment are further improved. The optimized model can dynamically fit the complex behaviors of the dummy, such as linear acceleration and deceleration, curvilinear motion, and random motion. It adjusts the state prediction direction through velocity feedback and filters out environmental clutter from the radar point cloud and measurement noise from the thermal imager, making the aligned spatial point cloud data more consistent with the actual motion patterns of the dummy. This optimization ensures that the temporal consistency of the data remains stable under complex motion scenarios, providing stronger support for the accuracy of subsequent temperature compensation and the reliability of test results.

[0019] Preferably, the temperature sensing compensation model includes an input layer, which receives surface temperature distribution data, extracts a two-dimensional thermal radiation image from the surface temperature distribution data, and enhances the pixel values ​​in the two-dimensional thermal radiation image to achieve low-temperature environment adaptation, thereby obtaining an enhanced two-dimensional thermal radiation image; the formula for enhancing the low-temperature environment adaptation of pixel values ​​is: f =(f f_background) [1+α exp( β f_background)]; where f represents the pixel value in the two-dimensional thermal radiation image, f This represents the low-temperature environment adaptation enhancement result of pixel value f, where f_background represents the ambient temperature of the test environment, α represents the low-temperature sensitivity coefficient, and β represents the ambient temperature attenuation factor.

[0020] Beneficial Effects: The low-temperature environment adaptation enhancement formula in the input layer effectively solves the problem of nonlinear attenuation of thermal radiation signals under low nighttime temperatures. The formula incorporates parameters such as the test environment temperature and low-temperature sensitivity coefficient, dynamically correcting the pixel values ​​of the thermal radiation image. This significantly improves the contrast between the dummy target and the background, avoiding thermal image blurring and unclear contours caused by low temperatures. The enhanced two-dimensional thermal radiation image provides high signal-to-noise ratio data for subsequent thermal anomaly identification and feature extraction in the temperature-sensing compensation model, ensuring accurate capture of dummy thermal radiation features even in extreme low-temperature environments, laying a solid foundation for dynamic temperature compensation.

[0021] Preferably, the temperature-sensing compensation model includes a multi-scale fusion module. The multi-scale fusion module is used to receive images of the flexible dummy torso region and perform convolutional fusion processing on adjacent flexible dummy torso region images to obtain fusion features for each flexible dummy torso region. The convolutional fusion processing procedure includes: extracting the flexible dummy torso region image for which fusion features are to be calculated, and adjacent flexible dummy torso images of the extracted image; performing convolutional processing on the extracted images respectively to obtain image features; calculating the attention weights of the features of adjacent flexible dummy torso images using an attention mechanism; performing attention weighting on the features of adjacent flexible dummy torso images; and fusing the attention weighting result with the features of the flexible dummy torso region image for which fusion features are to be calculated to obtain fusion features.

[0022] Beneficial Effects: The multi-scale fusion module, employing residual convolution and an attention mechanism, addresses the discontinuity of thermal features caused by dummy motion. The module extracts torso region image features from the current frame and adjacent frames, uses an attention mechanism to weight and highlight key thermal information, and then fuses it with the current frame features. This approach preserves local thermal structure information while reflecting the temporal variation of thermal features. This design effectively avoids misjudgments of thermal anomalies caused by noise or motion blur in single-frame images, improves the accuracy of torso region fusion feature extraction, and provides precise and continuous feature support for subsequent adaptive thermal anomaly recognition.

[0023] Preferably, the temperature compensation model further includes an adaptive thermal anomaly identification module. This module receives the fused features of the flexible dummy's torso region, generates an adaptive thermal threshold, and calculates the number of elements in the fused features below the adaptive thermal threshold. If the number of elements exceeds a preset threshold, it indicates that a thermal anomaly exists in the flexible dummy's torso region. The formula for calculating the adaptive thermal threshold is: φ=φ0+η std (1 exp( timeτ));η=0.1 log( (maxTem) where φ represents the adaptive heat threshold, η represents the baseline heat threshold, the mean of all fused features is calculated to form a fused feature mean sequence, and std represents the standard deviation of the fused feature mean sequence. maxTem represents the absolute value of the maximum difference between the mean values ​​of the fused features, time represents the motion time of the flexible dummy, τ represents the time control parameter, and η represents the adaptive gain.

[0024] Beneficial Effects: The dynamic adaptive thermal threshold algorithm completely solves the core pain point of misjudging thermal anomalies. By combining parameters such as the standard deviation of the fusion feature mean sequence and the dummy's movement time, the algorithm adjusts the thermal threshold in real time, accurately distinguishing motion-induced changes in thermal image pixels from genuine thermal anomalies. By quantifying the number of elements below the threshold in the fusion features, precise localization of thermal anomalies is achieved, reducing the misjudgment rate and avoiding ineffective or missed compensation, thus ensuring the targeted and reliable nature of the temperature compensation strategy.

[0025] Preferably, the temperature compensation model includes a heating pipe flow rate regulation module. This module is used to obtain the number of flexible dummy torso regions with thermal anomalies. If this number is higher than a preset threshold, the average surface temperature collected by the thin-film temperature sensor deployed on the surface of the flexible dummy is extracted, and the flow rate of the heat-conducting oil in the inner thermal structure is calculated. Conversely, if the number of flexible dummy torso regions with thermal anomalies is not higher than the preset threshold, the heating pipe flow rate regulation module does not operate. The formula for calculating the heat-conducting oil flow rate is: P = ρ C S [(Tem Tem [3 / 2Len]; where P represents the regulating flow rate of the heat transfer oil, ρ represents the density of the heat transfer oil, C represents the specific heat capacity of the heat transfer oil, S represents the cross-sectional area of ​​the serpentine heating pipe, Len represents the length of the serpentine heating pipe, and Tem represents the average surface temperature collected by the thin-film temperature sensor. This indicates the preset standard temperature.

[0026] Beneficial Effects: Through the heating pipeline flow regulation module and calculation formula, precise on-demand temperature control of the flexible dummy is achieved. The module flexibly switches between local and overall compensation (heat transfer oil flow regulation) modes based on a threshold number of thermal anomaly areas. This avoids energy waste during localized thermal anomalies and rapidly increases the overall temperature when multiple areas experience thermal anomalies. The calculation formula, based on the physical properties of the heat transfer oil, pipeline parameters, and temperature differences, enables precise quantitative control of the flow rate, shortening the compensation response time, improving temperature control accuracy, effectively preventing temperature overshoot or lag, and ensuring that the dummy's thermal distribution always conforms to human body characteristics.

[0027] Beneficial effects of the first option This proposed method for nighttime AEB (Autonomous Emergency Braking) pedestrian testing utilizes a temperature-sensing coupled flexible pedestrian testing approach. By combining thermal sensing characteristics with behavioral simulation mechanisms, it significantly improves the simulation accuracy and scene reproduction capability of the dummy in low-temperature nighttime environments. Firstly, at the structural level, a three-layer composite thermal structure is employed, achieving both efficient heat conduction and heat retention. The outer layer material provides infrared radiation characteristics similar to a real human body, thereby improving the accuracy of infrared thermal imagers and making the test more closely resemble the actual thermal field characteristics of pedestrians. Secondly, the flexible dummy's surface is equipped with heating elements and thin-film temperature sensors, enabling regional temperature control and thermal compensation. Even under extreme cold conditions (e.g., below -10℃), the core surface temperature of the flexible dummy can be maintained at 36-38℃, ensuring stable detection of the dummy target by the AEB system sensors. Regarding behavioral simulation, the flexible dummy is equipped with a controllable movement device, supporting various movement modes such as straight lines, curves, acceleration / deceleration, and crossing roads. It can also simulate unpredictable pedestrian behaviors such as sudden stops or turns, effectively increasing the complexity and challenge of the test. With the aid of wireless remote control and preset programs, testers can set the dummy's movement speed, direction, and behavior patterns, further enriching the test scenarios. Overall, this solution not only fills the gap in existing nighttime AEB testing by addressing both thermal simulation and behavioral realism, but also constructs a controllable and repeatable complex testing platform through hardware and software collaboration, providing a stable, high-precision, and highly realistic solution for extreme testing of autonomous driving perception systems.

[0028] Meanwhile, the proposed temperature-sensing compensation model possesses high intelligence and dynamic thermal response capabilities, enabling precise temperature compensation for flexible dummies during nighttime AEB testing, thus improving test reliability and target realism. The model uses a time-aligned two-dimensional thermal radiation image as input, combining it with the YOLOv7-Thermal thermal image analysis module to perform multi-scale fusion processing on the torso region of the flexible dummy. It also introduces residual convolution and attention mechanisms to extract and enhance key thermal features while preserving local structural information. An innovative dynamic adaptive thermal threshold calculation mechanism is introduced, utilizing the standard deviation of temperature changes and a nonlinear time response function to achieve accurate thermal anomaly identification, effectively avoiding overcompensation and misjudgment. Through collaborative control of the local temperature-sensing compensation module and the heating pipeline flow regulation module, it not only supports local heating compensation based on thin-film temperature sensor data but also precisely regulates the flow rate of heat transfer oil in the serpentine pipeline based on a thermodynamic model. This method integrates image processing, dynamic threshold adjustment, and thermal conduction modeling, improving the response speed and temperature control accuracy of the compensation strategy. It ensures that the flexible dummy retains realistic thermal distribution characteristics under different environments, significantly improving the recognition accuracy and test consistency of the nighttime AEB system.

[0029] The second solution is a temperature-sensitive coupled pedestrian flexibility testing system for nighttime AEB pedestrian testing, including a data analysis module. The data analysis module includes an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it enables the electronic device to implement the temperature-sensitive coupled pedestrian flexibility testing method for nighttime AEB pedestrian testing according to the first solution.

[0030] Beneficial effects: Leveraging the hardware computing power and software execution capabilities of electronic devices, complex processes such as test environment adaptation, dummy behavior control, multi-source data acquisition and timing alignment, temperature compensation model calculation, and full-dimensional result quantification are efficiently completed, significantly reducing errors caused by human intervention and improving test efficiency and repeatability. The data analysis module can process and deeply mine test data in real time, ensuring the dynamic response speed of temperature compensation and the objectivity of result evaluation. It is also adaptable to different nighttime lighting, low temperature environments, and dummy movement modes, providing a stable, controllable, and highly realistic test platform for the AEB system. This not only fully realizes the technical advantages of the first solution but also meets the standardized and large-scale requirements of the intelligent driving field for the nighttime performance verification of the AEB system, providing accurate and reliable data support for system optimization and iteration. Attached Figure Description

[0031] Figure 1 This is a schematic diagram illustrating an implementation method for a temperature-sensing coupled pedestrian flexibility testing method for nighttime AEB pedestrian testing; Figure 2 This is a schematic diagram of a temperature-sensing coupled pedestrian flexibility testing system for nighttime AEB pedestrian testing, as shown in Example 3. Figure 3 This is a schematic diagram of the electronic device structure of a temperature-sensing coupled pedestrian flexibility testing system for nighttime AEB pedestrian testing, as shown in Example 3.

[0032] The reference numerals in the accompanying drawings include: Processor 101, input device 102, output device 103, memory 104, bus 105, computer program 1041. Detailed Implementation

[0033] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0034] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0035] Example 1 This embodiment provides a temperature-sensing coupled pedestrian flexibility testing method for nighttime AEB pedestrian testing, including: A: Set up a test environment for nighttime AEB pedestrian testing, and deploy flexible dummies within the test environment.

[0036] Specifically, in the nighttime AEB pedestrian test environment, if the test environment is deployed outdoors, it is considered nighttime with a light intensity of 5 lux or less. If the test environment is deployed indoors, it is a low-light environment with a light intensity of 10 lux or less. Regardless of whether it is indoors or outdoors, the vehicle driving surface includes asphalt roads, simple gravel roads, and roads where snow and ice have melted. Flexible dummy types include child dummies, adult dummies, and elderly dummies. To further simulate the visual characteristics of real pedestrians at night, the flexible dummies wear dark clothing or non-reflective materials.

[0037] A composite thermal structure is deployed on the flexible dummy, comprising an inner, middle, and outer thermal layer. The inner thermal layer consists of serpentine heating pipes encased in a high thermal conductivity silicone substrate, used to circulate heat-conducting oil and provide heat. The middle thermal layer is an aerogel insulation layer used to block external low-temperature interference and improve thermal stability. The outer thermal layer is an infrared-simulating coating made of human skin-like infrared material, used to provide human-like thermal radiation to the infrared thermal imager. The thermal conductivity of the silicone substrate is greater than 3 watts per meter per Kelvin, where Kelvin represents a unit temperature difference, and 3 watts per meter per Kelvin means that, assuming the thickness of the high thermal conductivity silicone substrate is 1 meter, each square meter of the high thermal conductivity silicone substrate transfers 3 W of heat per second.

[0038] This design, featuring a composite thermal structure on the flexible dummy, addresses the issue of thermal radiation distortion in low-temperature environments at night. The outer layer simulates human infrared characteristics to enhance sensor recognition, the middle layer blocks external low temperatures to maintain thermal stability, and the inner layer efficiently conducts heat to ensure uniform heat distribution, keeping the dummy's surface temperature stable at 36-38℃ (human-like temperature). Even at night, regardless of the season or extreme temperatures (below -20℃), this design allows for testing in various scenarios, ensuring the dummy's thermal stability and significantly improving test realism. Furthermore, the inner layer uses thermally conductive silicone (thermal conductivity > 3W / (m²)). The high thermal conductivity of K) can quickly transfer heat from the heat transfer oil. Combined with the middle insulation layer, it reduces heat loss, lowers the energy consumption and workload of the heating pipeline, extends the continuous testing time of the dummy, and avoids frequent shutdowns for heating, which would affect the testing efficiency.

[0039] B: Set the behavioral logic of the flexible dummy that conforms to the AEB pedestrian test.

[0040] Specifically, the behavioral logic of the flexible dummy includes linear acceleration and deceleration, curvilinear motion, and random motion. The flexible dummy, which performs linear acceleration and deceleration, is mounted on a movable linear guide rail. A motor drives the linear guide rail, controlling the dummy's speed, acceleration, and direction of movement. Linear acceleration and deceleration is used to simulate scenarios where pedestrians cross the road or suddenly appear.

[0041] A robotic chassis is deployed at the bottom of a flexible dummy that performs curvilinear and random movements. The robotic chassis is battery-powered and uses wireless remote control and a pre-programmed sequence to control the dummy's direction, speed, and path of movement. Specifically, the curvilinear and random movements are used to simulate unpredictable pedestrian behavior, such as suddenly running out of bounds or changing direction.

[0042] As seen in A and B, the highly thermally conductive silicone substrate deployed on the flexible dummy combines flexibility and thermal conductivity, conforming to the bending movements of the dummy's torso and limbs. The heating pipes move with the dummy without any breaks in heat conduction, ensuring a uniform surface temperature distribution during complex movements such as straight lines and curves, thus adapting to dynamic testing scenarios. Furthermore, by supporting linear acceleration and deceleration (simulating crossing a road) and curved motion combined with random motion (simulating unpredictable behavior), and using a sliding rail / robot chassis for controllable movement, the system can replicate the complex behavior of real pedestrians, increasing the complexity and challenge of testing scenarios and more comprehensively verifying the AEB system's response capabilities to sudden and irregular targets.

[0043] By combining the recreation of real pedestrian-environment interaction scenarios at night, the system accurately reproduces the dual authenticity of pedestrians' "thermal characteristics and motion state" at night. The system uses the shape and posture of the dummy, nighttime thermal radiation, and behavioral randomness as reference factors for AEB pedestrian testing, which greatly restores the test under the dynamic behavior of real pedestrians in real scenarios.

[0044] C: Deploy a test vehicle in the test environment. The test vehicle is equipped with sensor components to acquire test data of a flexible dummy in front of the test vehicle.

[0045] Specifically, the test vehicle was equipped with an AEB system, vehicle condition monitoring device, control system, and sensor components. The sensor components included an infrared thermal imager, millimeter-wave radar, and a communication module. Test data included surface temperature distribution data and spatial position data of the flexible dummy.

[0046] Infrared thermal imagers are used to perceive thermal radiation in front of the test vehicle in real time and convert the thermal radiation into temperature to obtain surface temperature distribution data of the flexible dummy in front of the vehicle. The surface temperature distribution data includes two-dimensional thermal radiation image data and image acquisition timestamp. A millimeter-wave radar is used to transmit radar signals in front of the test vehicle and receive echo signals. Spatial position data of a flexible dummy in front of the vehicle is extracted from the echo signals. The millimeter-wave radar is used for real-time position sensing in front of the test vehicle to obtain the spatial position data of the flexible dummy. This spatial position data includes spatial point cloud data and timestamps of point cloud data acquisition. The spatial point cloud data includes the flexible dummy's three-dimensional position coordinates and velocity. After aligning the timestamps of the point cloud data acquisition with those of the image acquisition, the two-dimensional thermal radiation image data, the flexible dummy's three-dimensional position coordinates, and the flexible dummy's three-dimensional position coordinates are then aligned.

[0047] Furthermore, a unified system master clock is set in the control system of the test vehicle, serving as the master clock for the AEB system, vehicle status detection device, control system, and sensor components. The system master clock periodically emits TTL level trigger signals and simultaneously connects to the external trigger interfaces of the infrared thermal imager and millimeter-wave radar. Upon detecting the trigger signal, the infrared thermal imager and millimeter-wave radar immediately acquire data and record the timestamp of the acquisition results as the master timestamp.

[0048] Temporal alignment of surface temperature distribution data and spatial location data, including: Spatial point cloud data is extracted from spatial location data as observation data. Kalman filtering is then used to perform temporal alignment of the spatial point cloud data. The temporal alignment process includes: Construct a state transition model and an observation model for a flexible dummy, and provide initial state data and initial covariance at the initial timestamp.

[0049] For the state data at the current timestamp, a state transition model based on timestamp information is used to perform state transition, obtaining state transition information. Then, a prediction stage and an update stage are executed sequentially. In the prediction stage, state prediction and covariance prediction are performed on the state transition information sequentially. In the update stage, the observation model is used to convert the observation data into observation information. Combining the state prediction results, covariance prediction results, and observation information, Kalman gain, state update, and covariance update are performed on the state transition information sequentially. The state update result is used as the alignment update result of the spatial point cloud data at the current timestamp, and the state update result and covariance update result are used in the subsequent prediction stage.

[0050] The alignment update result for each timestamp is extracted and concatenated with the timestamp to form the spatial location data aligned with the time sequence of the surface temperature distribution data. The surface temperature distribution data is assumed to be the time-series aligned data. This addresses the issue of data asynchrony among multiple sensors, improves data consistency and accuracy, provides a reliable data foundation for subsequent temperature compensation and test result calculations, and reduces test errors caused by time sequence deviations.

[0051] It can be seen that thermal imagers (which collect temperature distribution data) and radars (which collect spatial location data) inherently have timestamp discrepancies due to differences in hardware response speed and data transmission latency. The optimized Kalman filter triggers data acquisition via a unified system master clock, combined with the time interval in the state transition model. The parameters correct the timestamp error between the two types of data to within milliseconds, completely resolving the mismatch between the dummy's temperature and position data at the same moment, and avoiding logical deviations in subsequent temperature compensation based on past positions. Specifically, an optimized Kalman filter is used to align the timestamps of the thermal imager (temperature data) and radar (spatial position data), and the optimized state transition model incorporates speed control and noise suppression.

[0052] In one embodiment of this invention, the state transition model in the Kalman filtering process is optimized, and the optimized state transition model is calculated as follows:

[0053] ; ; ; in, This represents the alignment update result of the spatial point cloud data at timestamp t-1. Representing spatial point cloud data The three-dimensional position coordinates in Representing spatial point cloud data The speed in the middle, This represents the state transition matrix corresponding to timestamp t. Represents the L2 norm. Indicates speed control parameters; This represents the noise parameters at timestamp t-1. This indicates that the mean is 0 and the covariance is... Multidimensional Gaussian distribution, Indicates the time interval between adjacent timestamps. This represents the variance of acceleration of the flexible pedestrians used in the nighttime AEB pedestrian test. This represents the state transition information at timestamp t.

[0054] The optimized state transition model incorporates velocity control parameters and noise suppression terms. On one hand, it dynamically adjusts the state prediction direction through velocity feedback, ensuring that the temporal alignment of the spatial point cloud data closely matches the actual motion patterns of the dummy (such as position-velocity correlation during linear acceleration / deceleration and curvilinear motion). On the other hand, it filters environmental interference (such as road surface reflections and nighttime clutter) and temperature measurement noise from the thermal imager into the radar point cloud data using multidimensional Gaussian noise parameters. After alignment, the temporal consistency between temperature distribution data and spatial location data is significantly improved, ensuring that each set of temperature pixel values ​​accurately corresponds to the dummy's three-dimensional coordinates at the same moment.

[0055] D: Construct a temperature compensation model based on the thermal radiation characteristics of specific parts, deploy the temperature compensation model in a flexible dummy, use the temperature compensation model to receive test data, generate temperature compensation strategies for the characteristic parts of the flexible dummy, and perform real-time dynamic temperature compensation for the flexible dummy.

[0056] Specifically, a temperature compensation model is deployed within the composite thermal structure of the flexible dummy, and it receives time-aligned surface temperature distribution data. The model takes the time-aligned surface temperature distribution data as input and outputs a temperature compensation strategy for specific parts of the flexible dummy. The model includes an input layer, a thermal mapping module, a multi-scale fusion module, an adaptive thermal anomaly identification module, a specific part temperature compensation module, a heating pipe flow regulation module, and a strategy output module. In this embodiment, the specific parts include the head region, chest region, abdominal region, shoulder region, and pelvic region.

[0057] The input layer receives time-aligned surface temperature distribution data and extracts a two-dimensional thermal radiation image from the surface temperature distribution data. Low-temperature environment adaptation enhancement is applied to the pixel values ​​in the two-dimensional thermal radiation image to obtain an enhanced two-dimensional thermal radiation image. Furthermore, the pixel values ​​in the two-dimensional thermal radiation image represent the temperature at the image pixel location, where the temperature is in degrees Celsius.

[0058] From steps C and D, firstly, to realistically recreate the real scene, temperature compensation analysis is performed on specific areas of the dummy's torso. The head, with its clear outline and concentrated thermal radiation, is a core feature area for AEB system to identify pedestrians at night. The temperature compensation model needs to generate compensation strategies based on the temperature of specific parts of the dummy and the dummy's spatial position at the same time (e.g., when the dummy moves to a windward area, the temperature of the corresponding part needs targeted compensation). The time-aligned dataset can accurately correlate the 3D coordinates of specific parts such as the dummy's head with the 2D thermal radiation image pixel values ​​of specific parts such as the head area. This allows the multi-scale fusion module to accurately extract the spatiotemporal fusion features of specific areas within the same torso, and the adaptive thermal anomaly recognition module to accurately determine whether the change is due to movement or a genuine temperature anomaly based on synchronous data, avoiding misjudgments caused by asynchronous data (e.g., misjudging lower temperatures at different times as thermal anomalies).

[0059] Secondly, the chest / abdomen region is used to compensate for the uniformity of heat distribution in the core torso. The chest and abdomen are the most stable core areas of human body heat radiation and are also key areas for temperature compensation. Time-series aligned data can accurately match "three-dimensional position of the torso (such as slight swaying of the torso while walking)" with "chest / abdomen temperature distribution data." The multi-scale fusion module can focus on the continuity of thermal characteristics in this area, and the adaptive thermal anomaly identification module can accurately locate "local heat dissipation anomaly points" (such as a certain area of ​​the abdomen where heat dissipates too quickly due to clothing gaps). The heating pipe flow regulation module, based on synchronized position and temperature data, specifically adjusts the flow rate of heat transfer oil in this area to ensure that the chest / abdomen temperature is stable at 36-38℃, maintaining human-like core heat radiation characteristics and avoiding misjudging the target type by the AEB system due to uneven heat distribution in the torso.

[0060] Specifically, the formula for enhancing the low-temperature environment adaptability of pixel values ​​is as follows: ; in, Represents the pixel values ​​in a two-dimensional thermal radiation image. Represents pixel value The results of enhanced adaptation to low-temperature environments This indicates the ambient temperature of the test environment. Indicates the low-temperature sensitivity coefficient. This represents the ambient temperature decay factor.

[0061] The system receives time-aligned two-dimensional thermal radiation images and enhances pixel values ​​using a low-temperature environment adaptation enhancement formula, incorporating ambient temperature, a low-temperature sensitivity coefficient α, and an ambient temperature attenuation factor β. This addresses the issue of nonlinear attenuation of thermal radiation signals in low-temperature environments (below -20°C) at night. Pixel values ​​are dynamically corrected based on ambient temperature, improving the contrast between the dummy target and the background in the thermal image. This prevents blurring and indistinct outlines caused by low temperatures, providing high signal-to-noise ratio input data for subsequent thermal anomaly identification and significantly improving the clarity of thermal images at low temperatures.

[0062] The thermal image analysis module uses the YOLOv7-Thermal framework to divide the two-dimensional thermal radiation image into multiple flexible dummy torso region images, including the head, chest, abdomen, shoulders, and pelvis. By achieving precise part-level compensation similar to that of a real human, and by differentiating the thermal radiation characteristics of different parts (such as concentrated thermal radiation at the head), it avoids overall overheating or undercompensation due to thermal anomalies in one part, ensuring that the temperature of each part remains stable at 36-38℃ (human-like temperature).

[0063] A multi-scale fusion module receives images of the torso region of a flexible dummy and performs convolutional fusion processing on adjacent torso region images to obtain the fused features for each torso region. This addresses the problem of discontinuous thermal features caused by dummy movements (such as arm swinging and turning). Residual convolution preserves local thermal structure information, while an attention mechanism focuses on key thermal features, enabling the fused features to reflect the temporal patterns of thermal changes in specific areas. This avoids misjudgments of thermal anomalies caused by noise or motion blur in single-frame images, thus improving feature extraction accuracy.

[0064] In this embodiment, residual convolution is used for convolutional fusion processing. The convolutional fusion processing procedure includes: extracting the flexible dummy torso region image to be fused and the adjacent flexible dummy torso images of the extracted image; performing convolution processing on the extracted images to obtain image features; calculating the attention weights of the adjacent flexible dummy torso image features using an attention mechanism; performing attention weighting on the adjacent flexible dummy torso image features; and fusing the attention weighting result with the flexible dummy torso region image features to be fused to obtain the fused features.

[0065] The adaptive thermal anomaly detection module receives the fused features of the flexible dummy's torso region, generates an adaptive thermal threshold, and calculates the number of elements in the fused features that are below the adaptive thermal threshold. If the number of elements is higher than a preset threshold, it indicates that a thermal anomaly exists in the flexible dummy's torso region; conversely, if the number of elements is not higher than the preset threshold, it indicates that no thermal anomaly exists in the flexible dummy's torso region. The preset threshold is a conventional setting. In this embodiment, the fused features are in matrix form. The formula for calculating the adaptive thermal threshold is: ;

[0066] in, Indicates the adaptive thermal threshold. The baseline heat threshold is represented by the mean value of all fusion features, which are then used to form a fusion feature mean sequence. The standard deviation of the fused feature mean sequence is represented by... This represents the absolute value of the maximum difference between the means of the fused features. This indicates the motion time of the flexible dummy. Indicates time control parameters. This indicates adaptive gain. The dynamic threshold algorithm solves the problem of misjudgment of thermal anomalies. The dynamic threshold can adapt to the duration of dummy movement and the amplitude of temperature fluctuations, accurately distinguishing between changes in thermal image pixels caused by movement (such as thermal image stretching caused by pelvic displacement) and real thermal anomalies (such as excessive heat dissipation from the shoulder facing the wind), reducing the misjudgment rate and avoiding ineffective or missed compensation.

[0067] The specific area temperature compensation module is used to identify the flexible dummy torso area with thermal anomalies and generate temperature compensation instructions for that area.

[0068] The heating pipe flow regulation module is used to obtain the number of flexible dummy torso regions with thermal anomalies. If this number exceeds a preset region number threshold, the module extracts the average surface temperature collected by the thin-film temperature sensor deployed on the flexible dummy surface and calculates the flow rate of heat-conducting oil in the serpentine heating pipes of the inner thermal structure. Otherwise, the module does not operate. The preset region number threshold is a standard setting.

[0069] The formula for calculating the flow rate regulation of the heat transfer oil is as follows: ; Where P represents the regulating flow rate of the heat transfer oil. This indicates the density of the heat transfer oil. This indicates the specific heat capacity of the heat transfer oil, and S represents the cross-sectional area of ​​the serpentine heating pipe. Indicates the length of the serpentine heating pipe. This represents the average surface temperature collected by the thin-film temperature sensor. This indicates the preset standard temperature. The system precisely calculates and switches between compensation modes for accurate heat transfer oil flow, enabling precise temperature control on demand. In cases of localized thermal anomalies, only the corresponding heating element is activated, avoiding energy waste. When multiple areas experience thermal anomalies, the overall temperature is rapidly increased through heat transfer oil flow adjustment, shortening the response time. Simultaneously, it ensures a non-linear match between the heat transfer oil flow and the temperature difference, preventing temperature overshoot or compensation lag, thus improving temperature control accuracy.

[0070] The strategy output module is used to obtain the temperature compensation command for the torso area of ​​the flexible dummy with thermal anomalies, as well as the heat transfer oil adjustment flow rate output by the heating pipeline flow regulation module, as a local temperature compensation strategy for the flexible dummy.

[0071] The temperature-sensing compensation model is used to receive time-aligned surface temperature distribution data and generate a local temperature compensation strategy for the flexible dummy. This strategy performs real-time dynamic temperature compensation on the flexible dummy, including: When the test vehicle's temperature sensor detects a flexible dummy in front of it, and collects surface temperature distribution and spatial location data of the dummy, the time-aligned surface temperature distribution data is sent to the temperature compensation model of the dummy via the communication module. Simultaneously, when the dummy moves randomly (e.g., suddenly turns) or accelerates / decelerates in a straight line (e.g., crossing a road), the heat dissipation rate of specific parts of the dummy changes abruptly. The multi-scale fusion module tracks these heat dissipation changes using time-series features, the adaptive thermal anomaly identification module dynamically adjusts the threshold, and the compensation strategy adapts in real-time to the movement, avoiding thermal radiation distortion during motion.

[0072] A local temperature compensation strategy for the flexible dummy is calculated using a temperature-sensing compensation model. The temperature-sensing compensation command and the flow rate regulation of the heat transfer oil in the flexible dummy's torso region with thermal anomalies are extracted from the local temperature compensation strategy. The extracted temperature-sensing compensation command is used to control the micro PTC heating elements on the surface of the flexible dummy's torso region for heating. The flow rate regulation of the heat transfer oil is used to adjust the flow rate of the serpentine heating pipes in the inner thermal structure. If the local temperature compensation strategy does not have a heat transfer oil flow rate regulation, the flow rate of the serpentine heating pipes in the inner thermal structure is not adjusted.

[0073] In the temperature compensation strategy, the middle aerogel insulation layer deployed on the dummy reduces external low-temperature interference and heat loss, making the compensation strategy more efficient. Simultaneously, the outer human-skin-like infrared coating makes the enhanced thermal image after input layer more closely resemble the infrared characteristics of a real human body, improving the similarity between fused features and human thermal radiation, and making the compensation strategy more aligned with the AEB system's recognition logic. Furthermore, the inner high thermal conductivity silicone layer accelerates the conduction of compensation heat, ensuring that when thermal anomalies occur in specific areas (such as the shoulder), the compensation heat is evenly distributed within a short time, avoiding the occurrence of localized hot or cold spots.

[0074] As can be seen, the three-layer thermal structure in step A provides a stable thermal foundation, the multi-mode motion in step B simulates real pedestrian behavior, and the compensation model in step D achieves dynamic and precise temperature control. These three elements form a closed-loop test encompassing the scenario, the tested object, and the testing method. For example, in a low-temperature environment (step A), when a dummy crosses the road with random motion (step B), the middle aerogel insulation layer (A) reduces heat loss. Step D, through multi-scale fusion and adaptive recognition, tracks the dynamic heat dissipation of the head and shoulders in real time. Through the adjustment of the thermal oil flow and the coordinated compensation of local heating elements, it ensures that the dummy's heat radiation remains close to the human body during movement. This improves the realism of the AEB system's nighttime pedestrian detection, tracking, and braking decisions, thoroughly restoring the system's true performance in extreme scenarios. Combining the test data, the nighttime AEB pedestrian test results are calculated and recorded. These results include the similarity of the flexible dummy's test data, the consistency of the flexible dummy's motion trajectory, and the evaluation index results of the AEB pedestrian test.

[0075] E: Control the flexible dummy to move irregularly in the test environment. When the AEB system triggers braking, calculate and record the nighttime AEB pedestrian test results based on the test data. The AEB pedestrian test results include the similarity of the flexible dummy's test data, the consistency of the flexible dummy's motion trajectory, and the evaluation index results of the AEB pedestrian test.

[0076] Specifically, when the AEB system triggers braking, it combines the time-aligned surface temperature distribution data and spatial location data to calculate and record the similarity of the temperature distribution on the surface of the flexible dummy, the consistency of the flexible dummy's motion trajectory, and the evaluation index results of the AEB pedestrian test.

[0077] When the AEB system triggers braking, it acquires the temperature of each surface position coordinate of the flexible dummy in front of the test vehicle. The surface position coordinate temperature is extracted by a thin-film temperature sensor deployed on the surface of the flexible dummy, and the extracted temperature of each surface position coordinate is converted into a two-dimensional vector. The length and width of the two-dimensional vector are consistent with the length and width of the two-dimensional thermal radiation image in the surface temperature distribution data.

[0078] The cosine similarity between the two-dimensional vector and the two-dimensional thermal radiation image of the flexible dummy in front of the test vehicle when braking is triggered is calculated as the temperature distribution similarity of the flexible dummy's surface. The actual motion trajectory of the flexible dummy is determined using spatial location data, and the consistency between the actual and controlled motion trajectories is calculated as the consistency of the flexible dummy's motion trajectory. In this embodiment, the consistency between the actual and controlled motion trajectories is calculated using the reciprocal of the Euclidean distance. Specifically, if the temperature distribution similarity is lower than a preset similarity threshold, or the consistency of the flexible dummy's motion trajectory is lower than a preset consistency threshold, the test data is automatically marked as invalid, and the heating system self-check and the flexible dummy control program self-check process are triggered. Combining the time-aligned temperature distribution data and spatial location data, three types of results are calculated: test data similarity, motion trajectory consistency, and AEB core evaluation indicators, covering three dimensions: data authenticity, scene reproduction, and system performance.

[0079] Specifically, due to the limitations of sensors such as cameras and radar in existing tests, they are susceptible to interference in low-light and low-temperature environments (e.g., insufficient light for cameras, low sensitivity of radar to low-reflectivity targets). This makes it impossible to stably collect fine-grained data such as dummy thermal radiation and dynamic position; only the final result of whether a collision occurred or not can be captured. Therefore, current AEB tests focus primarily on the single evaluation of whether collision avoidance occurred. Temperature distribution similarity (cosine similarity calculation) verifies the realism of dummy thermal radiation simulation, motion trajectory consistency (reciprocal of Euclidean distance) verifies the accuracy of behavioral simulation, and AEB evaluation indicators (collision avoidance effect, braking response time, and collision speed reduction) directly reflect the core performance of the system, achieving full-chain quantification from testing process to results, resulting in more specific and clear outcomes.

[0080] The evaluation metrics for AEB pedestrian testing include whether the AEB system can avoid collisions with pedestrians in a nighttime environment, the time from when the AEB system detects a pedestrian to when it begins braking, and whether the AEB system can significantly reduce the collision speed after detecting a pedestrian. In this embodiment, the evaluation metrics for AEB pedestrian testing are collected and calculated using a vehicle state detection device deployed in the test vehicle and spatial location data.

[0081] As can be seen, the time-series alignment data in step C, including temperature and position synchronization, enables step E to accurately lock onto the dummy's state (temperature distribution, 3D position, and velocity) at the moment AEB triggers braking. This ensures the accuracy of collision risk assessment and braking effect judgment, preventing situations where the dummy has left the danger zone but is mistakenly judged as having successfully avoided a collision due to asynchronous data. Simultaneously, the optimized Kalman filter reduces noise interference from radar point cloud data, making trajectory consistency calculations in step E more accurate. It can identify minute trajectory deviations of the dummy, triggering timely self-correction and ensuring accurate reproduction of the test scenario. Furthermore, thermal imager data supports temperature similarity calculations, and radar data supports trajectory and velocity index calculations, providing multi-dimensional verification of the AEB system's closed-loop performance throughout the detection, tracking, and braking process. This avoids the biased evaluation caused by missing data from a single sensor, allowing multiple sensors to be used synergistically even with their inherent limitations, thus improving the accuracy of AEB testing.

[0082] In summary, firstly, this invention addresses the limitation of coupling between thermal simulation and motion simulation. Existing technologies either emphasize thermal simulation (fixed posture, good thermal stability but limited behavior) or motion simulation (flexible posture, but prone to distortion due to heat radiation), with the two mutually constraining each other. This embodiment decouples the thermal and motion structures: a three-layer composite thermal structure ensures thermal stability, linear guide rails and the robot chassis ensure motion flexibility, and a temperature-sensing compensation model dynamically adapts to heat loss during motion, achieving simultaneous optimization of thermal simulation realism and motion complexity, filling the gap in the industry for dual simulation of dynamic thermal response and complex behavior.

[0083] Second, it achieves a paradigm shift from passive testing to proactive prediction. Existing AEB testing follows a passive process: mannequin movement, system response, and result evaluation. This process fails to predict issues such as thermal anomalies and data deviations during testing. This embodiment addresses these issues by aligning sensor data timing (pre-correcting data deviations), providing real-time feedback on thermal anomalies from a temperature-sensing compensation model (pre-avoiding radiation distortion), and verifying the validity of test results (pre-removing invalid data). This transforms the testing process from passively receiving results to proactively avoiding errors, improving the reliability of test data and preventing misjudgments caused by equipment malfunctions or insufficient scene reconstruction.

[0084] Third, it offers one-stop testing across almost all test scenarios. Existing tests are limited by the dummy's movement patterns and environmental adaptability, merely recreating some single scenarios. This embodiment allows for dynamic adjustment of the dummy's movement patterns, temperature compensation strategies, and test environment parameters. This includes: the dummy supports three movement modes; the compensation strategy adapts to a wide temperature range from -20℃ to normal; the test road surface covers asphalt, gravel, and snow-melted roads; and it achieves full coverage of pedestrian types (elderly / middle-aged / young), dummy behavior patterns, and environmental conditions. The test scenarios have been transformed from singular to multi-dimensional, fully verifying the real performance of the AEB system under normal, extreme, and emergency scenarios, providing accurate extreme condition data for system optimization.

[0085] Fourth, it addresses the issue of asynchronous data during multi-sensor collaboration. By introducing an optimized Kalman filter as an "intermediary," and through unified master clock triggering and state transition model optimization, the timing deviation between the two types of data is corrected to the millisecond level. This achieves precise binding of temperature and location at the same moment and location, enabling temperature compensation to respond specifically to heat dissipation anomalies in specific areas (such as precise heating when the head is on the windward side). This improves compensation accuracy and solves the core problem of misjudgment caused by asynchronous data. It also allows multiple sensors to collaborate in testing, improving test accuracy.

[0086] Beneficial effects of this embodiment First, it recreates realistic and diverse test scenarios. Through the collaborative design of the test environment, flexible dummies, and behavioral logic, it solves the core problem of the disconnect between nighttime AEB test scenarios and real-world operating conditions. In terms of environment construction, it accurately recreates the complex driving environment at night. In terms of dummy design, it employs a non-reflective / dark outer layer and a three-layer composite thermal structure. The outer layer, a human-skin-like infrared coating, provides human-like thermal radiation; the middle layer, an aerogel insulation layer, blocks low-temperature interference; and the inner layer, a highly thermally conductive silicone layer, ensures uniform heat conduction, keeping the dummy's surface temperature stable within a human-like range of 36-38℃, maintaining thermal stability even at extreme low temperatures of -20℃. In terms of behavioral simulation, it uses three logics—linear acceleration / deceleration, curvilinear motion, and random motion—to recreate both the conventional scenario of pedestrians crossing the road and simulate unpredictable behaviors such as sudden turns and stops. This achieves dual realistic simulation of thermal characteristics and motion states, ensuring that the AEB system test is based on the same detection target as actual nighttime pedestrians, guaranteeing test effectiveness from the test preparation stage.

[0087] Secondly, the multi-sensor approach was addressed by optimizing the timing alignment of multiple sensors, thus establishing a high-quality foundation of test data and providing reliable support for subsequent compensation and evaluation. A complementary acquisition scheme using infrared thermal imagers and millimeter-wave radar was employed. The thermal imager captured the surface temperature distribution and two-dimensional thermal radiation images of the dummy, while the radar acquired three-dimensional spatial data such as position coordinates and velocity. These two types of data were recorded with timestamps, compensating for the performance limitations of single sensors in low-light conditions at night. An optimized Kalman filter was used for timing alignment. By constructing state transition and observation models and incorporating velocity control parameters, noise suppression terms, and acceleration variance, the timestamp error between temperature and position data was corrected to within milliseconds. This ensured precise matching of temperature and spatial position of specific parts of the dummy at the same time, avoiding compensation misjudgments and evaluation biases caused by asynchronous data. Simultaneously, a unified system master clock triggered acquisition, further improving data synchronization and consistency, laying a solid foundation for subsequent accurate analysis.

[0088] Next, the temperature compensation model overcomes the limitations of overall dummy heating, achieving dynamic and precise temperature control for specific parts of the flexible dummy. The model takes time-aligned thermal radiation images as input and dynamically corrects pixel values ​​to improve thermal image contrast at low temperatures using a low-temperature environment adaptation enhancement formula in the input layer. It utilizes the YOLOv7-Thermal framework to divide specific regions such as the head, chest, and shoulders, and combines a multi-scale fusion module with residual convolution and attention mechanisms to accurately extract thermal features from each region. An adaptive thermal anomaly recognition module generates dynamic thresholds to accurately distinguish between thermal image changes caused by motion and genuine thermal anomalies, reducing the false positive rate. Finally, through the coordinated control of local heating elements and heat transfer oil flow regulation, targeted heating is activated for corresponding parts when there are local thermal anomalies, and the heat transfer oil flow is accurately calculated using formulas for multi-region anomalies, achieving on-demand compensation with high temperature control accuracy. This ensures that the dummy maintains realistic thermal distribution characteristics even under complex motion and extreme low temperatures.

[0089] Then, breaking away from the limitations of existing tests that only focus on collision avoidance results, a comprehensive evaluation system was constructed, encompassing process verification and outcome assessment. Temperature distribution similarity (cosine similarity calculation) verifies the realism of the dummy's thermal radiation simulation, ensuring the effectiveness of the test objective. Trajectory consistency (reciprocal of Euclidean distance) verifies the accuracy of the dummy's behavioral simulation, avoiding test bias caused by "adapting to a fixed trajectory." Core AEB indicators (collision avoidance effect, braking response time, and collision speed reduction) comprehensively reflect the system's performance throughout the entire process of detection, tracking, decision-making, and braking. Even if complete collision avoidance is not achieved, the system's intervention effect can be quantified through indicators such as response time and speed reduction. Simultaneously, a data validity verification mechanism is implemented. If the similarity or trajectory consistency does not reach a threshold, the data is automatically marked as invalid and a device self-check is triggered, eliminating erroneous data caused by equipment failure or insufficient scene reproduction at the source, ensuring the objectivity, repeatability, and persuasiveness of the test results.

[0090] Finally, through collaborative hardware and software design, the efficiency and applicability of nighttime AEB testing have been significantly improved. The three-layer composite thermal structure of the flexible dummy reduces heat loss. The inner layer of high-efficiency thermally conductive silicone and the middle insulation layer work together to reduce the energy consumption and workload of the heating pipes, extend the continuous testing time of the dummy, and avoid frequent shutdowns for heating that could affect the testing progress. The linear slide rails at the bottom of the dummy and the robot chassis support wireless remote control and preset program control. Testers can flexibly set the movement speed, direction, and mode, and quickly switch between different test scenarios without complex adjustments. The testing environment covers a wide temperature range from extreme low temperatures to normal temperatures. The dummy types include children, adults, and the elderly, and the behavioral logic covers various pedestrian characteristics, making the testing unrestricted by region, season, or crowd. It can meet the diverse testing needs of different AEB systems and provide a stable, controllable, and highly realistic solution for the extreme testing of autonomous driving perception systems.

[0091] Example 2 Unlike the previous embodiments, this embodiment provides a temperature-sensing coupled pedestrian flexibility testing system for nighttime AEB pedestrian testing, such as... Figure 2 As shown, a temperature-sensitive coupled pedestrian flexibility testing system for nighttime AEB pedestrian testing includes a data analysis module, which includes an electronic device. The electronic device includes a memory, a processor 101, and a computer program 1041 stored in the memory and executable on the processor. When the processor 101 executes the computer program, it enables the electronic device to implement a temperature-sensitive coupled pedestrian flexibility testing method for nighttime AEB pedestrian testing, as described in any of the above embodiments.

[0092] Specifically, such as Figure 3 As shown, the electronic device may include: one or more processors 101, one or more input devices 102, one or more output devices 103, one or more memories 104, and a computer program stored in the memory 104 and executable on the processor. The processor 101, input devices 102, output devices 103, and memory 104 are interconnected via a bus 105. The memory 104 stores the computer program 1041, which includes program instructions. The processor 101 is configured to invoke the program instructions to execute the method steps described in the above method embodiments.

[0093] It should be understood that, in this embodiment, the processor 101 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0094] Input device 102 may include a keyboard, etc., and output device 103 may include a display (LCD, etc.), a speaker, etc.

[0095] The memory 104 may include read-only memory and random access memory, and provides instructions and data to the processor 101. A portion of the memory 104 may also include non-volatile random access memory. For example, the memory 104 may also store device type information.

[0096] In specific implementations, the processor 101, input device 102, and output device 103 described in the embodiments of the present invention can execute the implementation methods described in the relevant embodiments of the temperature-sensing coupled pedestrian flexibility testing method and system for nighttime AEB pedestrian testing provided by the embodiments of the present invention, which will not be repeated here.

[0097] It should be noted that for a more detailed description of the electronic device's workflow and a method for performing a temperature-sensing coupled pedestrian flexibility test for nighttime AEB pedestrian testing, please refer to the aforementioned method implementation section, which will not be repeated here.

[0098] Example 3 Unlike the previous embodiments, the memory described in this embodiment should be interpreted broadly. It can be not only a hardware component in a computer system used for temporary data storage, but also a physical medium capable of storing digital information and being read by a computer. These media can be permanent or temporary, including but not limited to hard disks and solid-state drives.

[0099] Specifically, the memory can be an internal storage unit of the electronic device described in any of the embodiments, such as a system hard drive or memory. The memory can also be an external storage device of the system, such as a plug-in hard drive, SmartMediaCard (SMC), Secure Digital (SD) card, FlashCard, etc., equipped on the system. Furthermore, the memory can include both internal storage units and external storage devices. The memory is used to store the computer program and other programs and data required by the system. The memory can also be used to temporarily store data that has been output or will be output.

[0100] Storage devices include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media that can store program code.

[0101] Numerous specific details are set forth in this specification. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, systems, and techniques have not been shown in detail so as not to obscure the understanding of this specification. In the description of this specification, references to the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., mean that a specific feature, method, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this specification.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A warm body coupling pedestrian flexible test method for night AEB pedestrian test, characterized in that, The application relates to a night AEB pedestrian test method and a flexible dummy. The application comprises the following steps: A: setting a test environment for the night AEB pedestrian test, and deploying a flexible dummy in the test environment; B: setting the behavior logic of the flexible dummy in the AEB pedestrian test; C: deploying a test vehicle in the test environment, the test vehicle being provided with a sensor assembly, and obtaining test data of the flexible dummy in front of the test vehicle by the sensor assembly; wherein the test data comprises surface temperature distribution data and spatial position data of the flexible dummy; D: constructing a temperature compensation model based on the thermal radiation characteristics of specific parts, deploying the temperature compensation model in the flexible dummy, receiving the test data by the temperature compensation model, and generating a temperature compensation strategy for the specific parts of the flexible dummy to realize real-time dynamic temperature compensation of the flexible dummy; 2. The thermally coupled flexible pedestrian test method for night AEB pedestrian testing of claim 1, wherein, E: controlling the flexible dummy to perform irregular motion in the test environment, and calculating and recording the night AEB pedestrian test results when the AEB system triggers braking, wherein the AEB pedestrian test results comprise the similarity of the test data of the flexible dummy, the motion track consistency of the flexible dummy and the evaluation index results of the AEB pedestrian test. The outermost layer of the flexible dummy is made of non-reflective material or dark clothes; the composite thermal structure comprises an inner thermal structure, a middle thermal structure and an outer thermal structure in sequence; 3. The thermally coupled flexible pedestrian test method for night AEB pedestrian testing of claim 1, wherein, The inner thermal structure is a heating pipe wrapped by a heat-conducting silica gel base material, which is used for circulating heat-conducting oil and providing heat; the middle thermal structure is an aerogel thermal insulation layer, which is used for blocking the low-temperature interference outside the flexible dummy; the outer thermal structure is an infrared characteristic simulation coating, which is made of a human-skin-imitating infrared material and is used for providing human-like thermal radiation to an infrared thermal imager; the heat-conducting coefficient of the heat-conducting silica gel base material is greater than 3 W / m / K.

4. The thermally coupled flexible pedestrian test method of night AEB pedestrian testing of claim 1, wherein, The behavior logic of the flexible dummy comprises linear acceleration and deceleration motion, curve motion and random motion; wherein the linear acceleration and deceleration motion is used for simulating the scene of pedestrians crossing the road; the curve motion and the random motion are used for simulating unpredictable pedestrian behaviors. The sensor assembly comprises a thermal imager and a radar; the thermal imager is used for realizing real-time thermal radiation sensing in front of the test vehicle, converting the thermal radiation into temperature, and obtaining the surface temperature distribution data of the flexible dummy in front of the test vehicle; the surface temperature distribution data comprises two-dimensional thermal radiation image data and a time stamp of image acquisition; 5. The thermally coupled flexible pedestrian test method of night AEB pedestrian testing of claim 4, wherein, The radar is used for realizing real-time position sensing in front of the test vehicle, and obtaining the spatial position data of the flexible dummy in front of the test vehicle; the spatial position data comprises spatial point cloud data and a time stamp of point cloud data acquisition, wherein the spatial point cloud data comprises three-dimensional position coordinates and speed of the flexible dummy. The surface temperature distribution data and the spatial position data are time-aligned, comprising the following steps: extracting the spatial point cloud data in the spatial position data as observation data, and time-aligning the spatial point cloud data by using Kalman filtering; the time-alignment process comprises the following steps: constructing a state transition model and an observation model of the flexible dummy, and giving initial state data and initial covariance under an initial time stamp; For the state data at the current timestamp, state transition is performed using a state transition model based on timestamp information to obtain state transition information, and a prediction stage and an update stage are sequentially executed, wherein the prediction stage sequentially performs state prediction and covariance prediction on the state transition information, the update stage converts the observation data into observation information using an observation model, and combines the state prediction result, the covariance prediction result and the observation information to sequentially perform Kalman gain, state update and covariance update on the state transition information, and the state update result is taken as the alignment update result of the spatial point cloud data at the current timestamp, and the state update result and the covariance update result are used for subsequent prediction stages; The alignment update result at each timestamp is extracted, and the alignment update result is spliced with the timestamp as the spatial position data after time sequence alignment with the surface temperature distribution data, and the surface temperature distribution data is taken as the data after time sequence alignment by default.

6. The thermally coupled flexible pedestrian test method of night AEB pedestrian testing of claim 5, wherein, The state transition model in the Kalman filtering process is optimized, and the calculation method of the optimized state transition model is: ； ; ； wherein, denotes the alignment update result of the spatial point cloud data at the time stamp t-1, denotes the spatial point cloud data in the three-dimensional position coordinates, denotes the velocity of the spatial point cloud data in the three-dimensional position coordinates, denotes the state transition matrix corresponding to the time stamp t, denotes the L2 norm, denotes the velocity control parameter; denotes a noise parameter at time stamp t-1, denotes a multi-dimensional Gaussian distribution with mean 0 and covariance denotes a multi-dimensional Gaussian distribution with mean 0 and covariance denotes a time interval between adjacent time stamps, denotes the acceleration variance of a set flexible pedestrian in the night AEB pedestrian test; denotes state transition information at time stamp t.

7. The thermally coupled flexible pedestrian test method of night AEB pedestrian testing of claim 1, wherein, The temperature sensing compensation model includes an input layer, which is used to receive the surface temperature distribution data and extract a two-dimensional thermal radiation image from the surface temperature distribution data, and perform low-temperature environment adaptive enhancement on the pixel value in the two-dimensional thermal radiation image to obtain an enhanced two-dimensional thermal radiation image. The low-temperature environment adaptive enhancement formula of the pixel value is: ; wherein, represents a pixel value in a two-dimensional thermal radiation image, represents a pixel value a low-temperature environment adaptation enhancement result, represents an ambient temperature of a test environment, represents a low-temperature sensitivity coefficient, represents an ambient temperature decay factor.

8. The thermally coupled flexible pedestrian test method of night AEB pedestrian testing of claim 1, wherein, The temperature sensing compensation model includes a multi-scale fusion module, which is used to receive the flexible dummy torso region image and perform convolution fusion processing on adjacent flexible dummy torso region images to obtain the fusion feature of each flexible dummy torso region. The convolution fusion processing flow includes: extracting the flexible dummy torso region image to be calculated and the adjacent flexible dummy torso image of the extracted image, respectively performing convolution processing on the extracted images to obtain image features, calculating the attention weight of the adjacent flexible dummy torso image features using an attention mechanism, attention weighting the adjacent flexible dummy torso image features, and fusing the attention weighting result with the flexible dummy torso region image features to be calculated to obtain the fusion feature.

9. The thermally coupled flexible pedestrian test method of night AEB pedestrian testing of claim 8, wherein, The temperature sensing compensation model also includes an adaptive heat anomaly identification module, which is used to receive the fusion feature of the flexible dummy torso region and generate an adaptive heat threshold, calculate the number of elements in the fusion feature that are lower than the adaptive heat threshold, and if the number of elements is higher than a preset number threshold, it indicates that the flexible dummy torso region has a heat anomaly. The calculation formula of the adaptive heat threshold is: φ = φ0+ η std (1 exp( timeτ)) η = 0.1 log( maxTem) wherein φ represents an adaptive thermal threshold, η represents a reference thermal threshold, the mean of all fusion features is calculated to form a fusion feature mean sequence, std represents a standard deviation of the fusion feature mean sequence, maxTem represents an absolute value of a maximum difference of the fusion feature mean, time represents a motion time of the flexible dummy, τ represents a time control parameter, and η represents an adaptive gain.

10. The thermally coupled flexible pedestrian test method of night AEB pedestrian testing of claim 2, wherein, The temperature sensing compensation model includes a heating pipe flow regulation module, which is used to obtain the number of flexible dummy torso regions with heat anomalies, and if the number is higher than a preset region number threshold, the average surface temperature of the thin film temperature sensor deployed on the surface of the flexible dummy is extracted, and the heat conduction oil regulation flow in the inner heat structure is calculated; Conversely, the number of flexible dummy torso regions with heat anomalies is not higher than the preset region number threshold, and the heating pipe flow regulation module does not work; The calculation formula of the heat conduction oil regulation flow is: P = p C S [(Tem Tem )3 / 2Len]; wherein P represents the flow rate of the heat transfer oil, p represents the density of the heat transfer oil, C represents the specific heat capacity of the heat transfer oil, S represents the cross-sectional area of the serpentine heating pipe, Len represents the length of the serpentine heating pipe, Tem represents the average surface temperature collected by the thin film temperature sensor, and Tem represents the preset standard temperature.

Images