A rear vehicle risk early warning method and vehicle

Abstract

This application discloses a rear vehicle risk warning method and vehicle. It acquires a first signal from a rear-facing photosensitive sensor on the rearview mirror and a second signal from a forward-facing photosensitive sensor on the rearview mirror. Based on the first and second signals, it determines a change characteristic representing the change in the headlights of a rear vehicle. According to the change characteristic, it determines the risk level of the rear vehicle. Based on the risk level, it executes a corresponding warning operation. This application directly reuses existing forward and rear-facing photosensitive sensors on the vehicle's rearview mirror, eliminating the need for additional hardware costs. By jointly processing the signals from the two sensors, it effectively eliminates ambient light interference and utilizes the good penetrability of light signals under adverse conditions such as rain, fog, and nighttime to achieve reliable perception of the approaching risk of rear vehicles. Furthermore, it executes graded warnings based on the risk level, achieving all-weather rear collision warnings at low cost.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and vehicle for early warning of rear vehicle risks. Background Technology

[0002] Rear safety warning is a crucial function in a vehicle's active safety system. It aims to identify the risk of high-speed approach from vehicles directly behind or to the side, and warn the driver before a potential rear-end collision occurs, thus gaining valuable reaction time. This function is particularly important in medium-to-high-speed driving scenarios such as highways and urban expressways, effectively reducing the incidence of rear-end collisions caused by factors such as following vehicles not maintaining a safe distance, driver distraction, or obstructed visibility in inclement weather. Summary of the Invention

[0003] Based on the above problems, this application provides a rear vehicle risk warning method and vehicle.

[0004] The embodiments of this application disclose the following technical solutions: The first aspect of this application provides a method for warning of rear vehicle risks, including: Acquire the first signal from the rearward photosensitive sensor on the interior rearview mirror and the second signal from the forward photosensitive sensor on the interior rearview mirror; Based on the first signal and the second signal, determine the change characteristics used to characterize the changes in the lights of the vehicle behind; Based on the aforementioned changes, the risk level of the vehicles behind is determined; Based on the risk level of the vehicles behind, execute the corresponding warning operation.

[0005] In one possible implementation, determining the changing features characterizing the changes in the headlights of following vehicles based on the first signal and the second signal includes: Based on the second signal, ambient light compensation is performed on the first signal to obtain a third signal characterizing the contribution of the lights from vehicles behind. The dynamic change curve of the light intensity of the rear vehicle lights is extracted based on the third signal. Based on the dynamic change curve of light intensity, the change characteristics used to characterize the changes in the headlights of vehicles behind are determined.

[0006] In one possible implementation, the step of extracting the dynamic change curve of the light intensity of the rear vehicle lights based on the third signal includes: The third signal is subjected to bandpass filtering to obtain the filtered fourth signal; When the amplitude of the fourth signal is greater than or equal to the first amplitude threshold, the current moment is marked as the starting point of the optical event; Starting from the origin of the optical event, the amplitude of the fourth signal is recorded over time until the amplitude of the fourth signal is lower than the first amplitude threshold. The recorded time series data is then used as the dynamic change curve of the light intensity.

[0007] In one possible implementation, the light intensity dynamic change curve is characterized by at least one of the following: The signal peak change rate is the maximum value of the first derivative of the dynamic change curve of the light intensity, used to characterize the relative speed of the vehicle behind. Standardized rise time, which is the time required for the light intensity dynamic change curve to rise from the first percentage of the peak to the second percentage, is used to characterize the approach speed or initial distance of the following vehicle. The signal energy curve integral is the integral value of the dynamic change curve of light intensity on the time axis, which is used to characterize the light source intensity or proximity of the vehicle behind. Frequency domain modulation features, which are obtained by performing a short-time Fourier transform on the dynamic change curve of light intensity, include pulse features for identifying large vehicles or flicker features for identifying preset light source types. The double-peak detection result is used to determine whether there are two adjacent peaks in the curve, which is used to distinguish between vehicles with two lights and vehicles with one light. The peak platform judgment result is used to determine whether the curve maintains a flat range after reaching its peak, which is used to distinguish whether the following vehicles are following at a constant speed.

[0008] In one possible implementation, before determining the risk level of the following vehicle based on the changing characteristics, the method further includes: Obtain vehicle status information, which includes at least one of vehicle speed, vehicle acceleration, turn signal status, and weather information; The process of determining the risk level of vehicles behind based on the aforementioned changes includes: The change features and the vehicle state information are input into a pre-trained machine learning model so that the machine learning model outputs the risk level of the vehicle behind. The pre-trained machine learning model is built based on a lightweight gradient boosting decision tree or support vector machine. It is obtained by collecting real vehicle data and simulation data in different scenarios, using the change features and the vehicle state information as input feature vectors, and using the corresponding risk level, relative speed estimate and distance estimate as labels for supervised learning training.

[0009] In one possible implementation, the output of the pre-trained machine learning model further includes a relative speed estimate and a distance estimate; the step of performing a corresponding warning operation based on the risk level of the vehicle behind includes: When the risk level is Level 1 or Level 2 and the estimated collision time is greater than the first time threshold, a preparatory warning operation is performed. The estimated collision time is the quotient of the distance estimate and the relative speed estimate. The preparatory warning operation includes a visual cue for the first display effect when the driver's line of sight is detected to be directed toward the interior rearview mirror. When the risk level is level three and the expected collision time is between the second time threshold and the first time threshold, a first-level warning operation is executed, which includes a visual cue with a second display effect. When the risk level is level four and the expected collision time is between the third time threshold and the second time threshold, a second-level warning operation is executed. The second-level warning operation includes an auditory prompt with a first auditory effect and a visual prompt with a third display effect. When the risk level is level 5 and the estimated collision time is less than the third time threshold, a level 3 warning operation is executed. The level 3 warning operation includes visual cues with a fourth display effect, auditory cues with a second auditory effect, tactile cues, and vehicle pre-control actions. The first time threshold is greater than the second time threshold, and the second time threshold is greater than the third time threshold.

[0010] In one possible implementation, the step of performing ambient light compensation on the first signal based on the second signal to obtain a third signal characterizing the contribution of rear vehicle headlights includes: Subtracting the product of the second signal and the calibration coefficient from the first signal yields the third signal characterizing the contribution of the rear vehicle's headlights.

[0011] One possible implementation also includes: Obtain user feedback on the warning operation; Based on the feedback information, the user's personalized warning threshold is adjusted within a safe range.

[0012] One possible implementation also includes: When the signal amplitude of the rearward photosensitive sensor is continuously lower than the second amplitude threshold for a duration exceeding the first preset duration, and the signal amplitude of the forward photosensitive sensor is within the preset normal range, a cleaning reminder message is generated. The cleaning reminder message is used to remind the driver to clean the mirror surface of the rearview mirror or the light-sensing window of the rearward photosensitive sensor. When the anti-glare function of the rearview mirror is detected to be off, or when the photosensitive surface of the rear-facing photosensitive sensor is blocked by a foreign object or the light path is obstructed, a warning message indicating that the warning function is unavailable is generated.

[0013] A second aspect of this application provides a vehicle including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned rear vehicle risk warning method.

[0014] Compared with the prior art, this application has the following beneficial effects: This application acquires the first signal from the rear-facing light sensor and the second signal from the front-facing light sensor on the rearview mirror. These two sensors are originally used for the automatic anti-glare function and are existing hardware on the vehicle, thus eliminating the need for any additional dedicated sensors and avoiding the high costs associated with adding radar or cameras. Secondly, by jointly processing the first signal from the rear-facing light sensor and the second signal from the front-facing light sensor, this application effectively eliminates interference from changes in ambient light on the detection of following vehicle lights, ensuring accurate extraction of the changing characteristics of following vehicle lights even in complex lighting scenarios such as nighttime, tunnel entrances and exits, and strong glare or backlighting. Since light signals have significantly stronger penetrating power than radar waves in adverse weather conditions such as heavy rain and dense fog, as long as the following vehicle has its headlights on, the changes in its lights can be stably captured by the sensor, thus filling the technical blind spots of millimeter-wave radar performance degradation in adverse weather and camera recognition limitations in low light conditions. Finally, this application determines the risk level of the following vehicle based on the extracted changing characteristics and executes a warning operation matching the risk level. Compared to existing solutions that rely on dedicated sensors, this application achieves early perception and graded warning of the risk of high-speed rear-end collisions at extremely low hardware cost. It can reliably ensure the effectiveness of the rear collision warning function under different weather and lighting conditions, and has good economy and all-weather adaptability. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating a rear vehicle risk warning method provided in this application embodiment; Figure 2 This is a structural diagram of a rear vehicle risk warning system provided in an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0018] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0019] As mentioned earlier, existing rear-view safety systems primarily rely on sensors such as millimeter-wave radar, ultrasonic radar, and rearview cameras. However, these sensors have significant technical limitations in application. First, while millimeter-wave radar excels at ranging and speed measurement, its wide beam limits its ability to distinguish the directional positions of targets with small distances or low speed differences in the same lane directly behind the vehicle. Furthermore, in adverse weather conditions such as heavy rain and dense fog, radar signals are prone to attenuation and generate a large amount of clutter, resulting in a shortened detection range and a severe decrease in target recognition reliability, making it difficult to reliably identify the risk of a high-speed rear-end collision. Ultrasonic radar uses the time-of-flight of sound waves for ranging, with a typical effective range of less than 5 meters, making it only suitable for low-speed parking scenarios and unable to provide effective warnings at high speeds. Second, rearview cameras rely on visible light imaging. In scenarios with a large dynamic range, such as at night, tunnel entrances and exits, and strong light or backlight, the image signal-to-noise ratio drops sharply, severely limiting recognition performance. At the same time, image processing algorithms have a high computational load, placing stringent demands on onboard computing resources. Third, enhancing warning capabilities by adding dedicated sensors, such as long-range millimeter-wave radar or infrared cameras, would significantly increase the overall vehicle hardware cost, hindering its adoption in low- and mid-range models. Therefore, how to reliably and promptly detect the risk of a rear-end collision from a vehicle approaching at high speed under adverse weather and complex lighting conditions without increasing hardware costs has become a pressing technical problem in this field.

[0020] To address the aforementioned issues, this application embodiment specifically addresses the performance degradation of existing radars in adverse weather conditions. Instead of relying on radar waves that are easily attenuated by rain and fog, this application utilizes optical signals to sense vehicles behind. Light's penetrating power in heavy rain and dense fog is significantly superior to millimeter-wave radar waves. As long as a vehicle behind has its headlights on, its light can penetrate the rain and fog and be stably captured by the sensor, thus filling the performance blind spot of radar in adverse weather conditions.

[0021] To address the limitations of cameras at night or in strong light, this application uses the rear-facing and forward-facing light sensors inherent in the rearview mirror. By using ambient light compensation technology, it eliminates interference from ambient light such as streetlights, tunnels, and sunlight, obtaining a pure signal that only reflects the light contribution of vehicles behind. This requires no complex image processing, has a very low computational load, and is even more sensitive in scenarios with higher light contrast at night.

[0022] To address the short detection range of ultrasonic radar, this application utilizes the inverse square law of illuminance. By analyzing the physical laws governing the change of light intensity with distance, it can capture the headlight signals of vehicles behind from over 100 meters away, achieving long-range early warning and fully covering high-speed driving scenarios. To address the high cost of adding dedicated sensors, this application directly reuses the two front and rear photosensitive sensors from the vehicle's existing automatic anti-glare rearview mirror system. Without increasing hardware costs, only a software algorithm upgrade is needed to add a rear collision warning function to the vehicle, demonstrating extremely high economic efficiency.

[0023] Furthermore, this application extracts the dynamic change curve of the light intensity of following vehicle lights through high-frequency sampling, and obtains multi-dimensional features such as peak change rate, standardized rise time, frequency domain modulation characteristics, and bimodal structure from the curve. Combined with contextual information such as vehicle speed, acceleration, and turn signal status, a lightweight machine learning model is used to comprehensively assess the risk level. Finally, based on the risk level, a graded response is executed, ranging from non-visual cues to multimodal warnings and vehicle pre-control. Thus, this application achieves all-weather, long-range, and highly reliable early warning of rear-end collision risks at extremely low cost.

[0024] It should be noted that the rear vehicle risk warning method, product, and medium provided in this application can be applied to the field of data processing technology. The above are merely examples and do not limit the application field of the rear vehicle risk warning method, product, and medium provided in this application. Furthermore, the embodiments of this application may not limit the executing entity of the control; for example, the rear vehicle risk warning method in the embodiments of this application can be executed by a controller in the vehicle. In the technical solution of this application, the system mentioned refers to a control unit or control module integrated inside the vehicle to realize the rear vehicle approach warning function. The vehicle can refer to a passenger car or commercial vehicle with an automatic anti-glare function for the rearview mirror, and is the overall transportation platform on which the embodiments of this application are attached. The vehicle controller refers to various electronic control units in the vehicle's electronic and electrical architecture, specifically including but not limited to the body domain controller, cockpit domain controller, intelligent driving domain controller, or the microcontroller of the rearview mirror module itself. The raw signals from the rear and front light sensors can first be connected to the rearview mirror control module, which itself contains a microcontroller for basic control of the automatic anti-glare function. The technical solution of this application can be integrated as a software function module into the microcontroller of the rearview mirror control module, or it can run on a domain controller with higher computing power, and read the raw sampling data of the photosensitive sensor from the rearview mirror control module through the controller local area network or local interconnection network.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0026] The following embodiment illustrates a rear vehicle risk warning method provided in this application. See also... Figure 1 ,Should Figure 1 A flowchart of a rear vehicle risk warning method provided in this application embodiment, the method including: S101. Acquire the first signal from the rearward photosensitive sensor on the interior rearview mirror and the second signal from the forward photosensitive sensor on the interior rearview mirror.

[0027] In this embodiment, two types of raw signals are first obtained from the vehicle's rearview mirror control module. The rearview mirror is an optical reflector installed above the windshield inside the vehicle's driver's compartment for the driver to observe road conditions directly behind. This rearview mirror integrates an automatic anti-glare function, therefore, two photosensitive sensors are mounted on the mirror. The first signal comes from the rear-facing photosensitive sensor on the rearview mirror, which is mounted facing the rear of the vehicle and is specifically used to sense the intensity of light from the headlights of vehicles behind that illuminate the photosensitive surface.

[0028] The system directly reads the raw analog voltage signal output by the sensor from the rearview mirror control module, denoted as v_raw(t), where t represents time. Alternatively, a high-resolution digital sequence processed by an analog-to-digital converter can be read. To accurately capture rapid transient changes in the intensity of rear vehicle headlights and avoid losing key features due to excessively long sampling intervals, the sampling frequency needs to be set to at least 100 Hz. The second signal comes from a forward-facing photosensitive sensor on the rearview mirror, which is mounted facing forward of the vehicle and senses the ambient light intensity in front of the vehicle, including sunlight, streetlights, tunnel lighting, and skylight on cloudy days. The system reads the sensor's output signal in real time using a front-to-back synchronous method, also at a sampling frequency of at least 100 Hz, denoted as v_front(t). The forward sensor signal serves as an ambient light reference, providing a baseline for subsequent elimination of ambient light interference in the rearward signal, thereby extracting a pure signal that only reflects the contribution of rear vehicle headlights. By synchronously acquiring signals from both photosensitive sensors at high frequency, this step lays a complete data foundation for subsequent signal compensation, feature extraction, and approximation judgment.

[0029] S102. Based on the first signal and the second signal, determine the change characteristics used to characterize the changes in the lights of the following vehicle.

[0030] In one possible implementation, based on the first signal and the second signal, determining the changing characteristics for characterizing the changes in the headlights of following vehicles includes the following steps A1-A3: Step A1: Perform ambient light compensation on the first signal based on the second signal to obtain a third signal characterizing the contribution of the headlights from vehicles behind.

[0031] After acquiring the first and second signals, this step requires performing ambient light compensation on the first signal to eliminate interference from changes in ambient light on the detection of headlights of vehicles behind. Ambient light compensation refers to using the ambient light signal collected by the forward-facing photosensor as a reference, subtracting the ambient light component from the original signal of the rear-facing photosensor, thereby extracting the effective signal contributed only by the headlights of vehicles behind.

[0032] In one possible implementation, the step of performing ambient light compensation on the first signal based on the second signal to obtain a third signal characterizing the contribution of rear vehicle lights includes: subtracting the product of the second signal and a calibration coefficient from the first signal to obtain the third signal characterizing the contribution of rear vehicle lights.

[0033] Ambient light compensation is achieved through the following calculation: The first signal is denoted as v_raw(t), the second signal as v_front(t), and the difference signal v_rear(t) is calculated as v_raw(t) - k × v_front(t). Here, t represents time, and k is a pre-calibrated coefficient. The calibration coefficient k is used to adjust the ratio of the forward sensor signal to the rear sensor's response to ambient light, so that under pure ambient light conditions without rear vehicle headlights, the compensated signal v_rear(t) is close to zero. The calibration coefficient k can be determined through offline calibration experiments, such as measuring the ratio of v_raw(t) to v_front(t) under uniform illumination without rear vehicles, or through a one-time calibration at the vehicle's factory. The v_rear(t) obtained after the above ambient light compensation is the third signal, which mainly reflects the actual light intensity contribution of the rear vehicle headlights to the rear photosensitive sensor, while interference caused by slow fluctuations or abrupt changes in ambient light is effectively suppressed. The quality of the third signal directly affects the purity of the subsequent dynamic light intensity change curve. Therefore, ambient light compensation is a key prerequisite for ensuring that this technical solution works accurately in complex lighting scenarios.

[0034] Step A2: Extract the dynamic change curve of the light intensity of the rear vehicle lights based on the third signal.

[0035] In one possible implementation, extracting the dynamic change curve of the light intensity of the rear vehicle lights based on the third signal includes: performing bandpass filtering on the third signal to obtain a filtered fourth signal; marking the current moment as the optical event start point when the amplitude of the fourth signal is greater than or equal to a first amplitude threshold; recording the change of the amplitude of the fourth signal over time from the optical event start point until the amplitude of the fourth signal is lower than the first amplitude threshold, and using the recorded time series data as the dynamic change curve of the light intensity.

[0036] After obtaining the third signal after ambient light compensation, this step requires extracting a dynamic curve from the signal that reflects the change in the intensity of the headlights of following vehicles over time. Since the third signal may still contain residual low-frequency fluctuations from ambient light and high-frequency noise generated by the sensor circuitry itself, filtering is necessary before event detection. In one possible implementation, bandpass filtering is performed on the third signal to obtain a filtered fourth signal. Bandpass filtering is a signal processing technique that allows only signal components within a specific frequency range to pass through, while suppressing signals below the lower limit and above the upper limit. The bandpass filter used in this application has a passband frequency range of 0.1 Hz to 10 Hz. This range is designed based on the characteristic frequency of the rate of change of light intensity caused by the approaching headlights of following vehicles falling within this range. Signals below 0.1 Hz mainly originate from slow ambient light fluctuations such as sunrise and sunset, and entering and exiting tunnels, and are considered low-frequency interference, which should be filtered out. Signals above 10 Hz mainly originate from high-frequency interference such as circuit thermal noise, analog-to-digital conversion quantization noise, and rapid wiper oscillation, and should also be suppressed. The fourth signal obtained after bandpass filtering has a significantly improved signal-to-noise ratio, laying the foundation for accurate detection of optical events in the future.

[0037] After filtering, the system needs to monitor the amplitude changes of the fourth signal in real time and determine the starting point of the optical event accordingly. This step sets a first amplitude threshold, which can be a pre-calibrated fixed value or an adaptive baseline threshold dynamically calculated based on historical signals. One way to calculate the adaptive baseline threshold is to calculate the mean or median of the fourth signal over a past period and add an offset to this value, thus obtaining a dynamic threshold that can slowly change with the ambient light reference. Regardless of the method used, this threshold distinguishes between normal background light fluctuations and valid signals caused by the lights of vehicles behind. When the amplitude of the fourth signal changes from below the first amplitude threshold to above or equal to the threshold, the system determines the current moment as the starting point of the optical event. The starting point of an optical event signifies that the lights of vehicles behind have begun to be effectively captured by the sensor, and their intensity has significantly exceeded the ambient background, marking the beginning of the approaching process of a vehicle behind.

[0038] Starting from the origin of the optical event, the system continuously records the change in amplitude of the fourth signal over time, recording one data point every sampling period, forming a series of numerical sequences arranged chronologically. This recording process continues until the amplitude of the fourth signal drops below the first amplitude threshold, at which point the system determines the optical event has ended and stops recording. Finally, all time-series data recorded from the origin to the end of the optical event is used as the dynamic intensity change curve of the headlights of following vehicles. The horizontal axis of this curve represents time, and the vertical axis represents the filtered signal amplitude, which is proportional to the actual light intensity of the headlights of following vehicles illuminating the rear-facing photosensitive sensor. The specific shape of the dynamic intensity change curve, such as the rate of rise, the peak value, the presence of multiple peaks, and whether a plateau is maintained after the peak, contains rich information about the relative speed, distance, vehicle type, and motion state of the following vehicles, serving as the basis for subsequent risk level assessment. Through the above filtering, threshold comparison, and event recording, this step successfully transforms the original photosensitive signal into a dynamic curve with clear physical meaning, preparing for feature extraction and risk assessment.

[0039] Step A3: Based on the light intensity dynamic change curve, determine the change characteristics used to characterize the changes in the headlights of vehicles behind.

[0040] After obtaining the dynamic change curve of light intensity, this step requires extracting multi-dimensional features from the curve that can reflect the motion state of the vehicle behind and the characteristics of the light source, and comprehensively judging the degree of danger of the approaching vehicle based on these features.

[0041] In one possible implementation, the shape of the dynamic change curve of light intensity contains rich physical information, which will be explained below from three aspects: time-domain motion characteristics, frequency-domain modulation characteristics, and signal morphology characteristics.

[0042] First, temporal motion characteristics. Temporal motion characteristics are directly extracted from the amplitude changes and time relationships of the curve, mainly including the signal peak rate of change, normalized rise time, and signal energy curve integral. The signal peak rate of change refers to the maximum value of the first derivative of the dynamic light intensity curve, i.e., the slope at the steepest point along the rising edge of the curve. Physically, this value is strongly correlated with the relative speed of vehicles behind: the faster the vehicle approaches, the greater the increase in light intensity received by the sensor per unit time, and the higher the peak rate of change. Therefore, this feature can be used to characterize the relative speed of vehicles behind. The normalized rise time refers to the time required for the curve to rise from the first percentage of the peak to the second percentage. In practical implementation, it is often taken as the time difference from 10% to 90% of the peak. The shorter the normalized rise time, the faster the vehicle approaches or the closer the initial distance; conversely, a longer normalized rise time indicates a slower approach or a greater distance. The signal energy curve integral refers to the integration of the dynamic light intensity curve on the time axis. The calculation formula is: The integral value of V_rear(t)dt represents the total light energy received by the sensor during the observation period. Within the same time window, a larger integral value usually means that the vehicle behind is closer to the vehicle, or that its headlights are brighter, such as when the high beams are on. Therefore, this integral value can be used to characterize the light intensity or proximity of the vehicle behind.

[0043] Second, frequency domain modulation characteristics. Frequency domain modulation characteristics are obtained by performing a short-time Fourier transform on the dynamic change curve of light intensity. This transform converts the time-domain signal to the frequency domain, thereby analyzing the specific frequency components present in the signal. Different vehicle models or different light source types will leave unique frequency imprints in the light intensity signal, including pulse characteristics for identifying large vehicles or flicker characteristics for identifying preset light source types. For example, when a large truck is traveling at high speed, due to its less streamlined body shape, aerodynamic effects will cause low-frequency swaying of the front or body. This swaying causes the headlight beam direction to oscillate periodically, thus generating a pulse characteristic of about 2 Hz in the sensor signal. Therefore, when a pulse frequency component of about 2 Hz is detected, the system can determine that the vehicle behind is a large vehicle. As another example, LED headlights using pulse width modulation dimming technology have a driving current that itself contains a high-frequency switching frequency, resulting in high-frequency flickering of the emitted light intensity; in addition, damaged lamp covers or dirty lenses will also scatter and modulate the light, forming high-frequency components. Therefore, when high-frequency flickering features above a certain frequency threshold are detected in the spectrum, the system can identify whether the vehicle behind is equipped with LED headlights or has abnormal headlight covers, which helps to distinguish between different vehicle models and light source types. Furthermore, if the dynamic change curve of light intensity exhibits a long-term stable and flat spectrum after reaching its peak, without obvious low-frequency fluctuations or high-frequency flickering, it indicates that the vehicle behind is likely stationary, such as a parked vehicle with its high beams on, rather than a moving target approaching the vehicle. This judgment can effectively suppress false triggering.

[0044] Third, signal morphological features. Signal morphological features focus on the geometric structure of the curve shape, mainly including bimodal detection results and peak plateau judgment results. Bimodal detection refers to judging whether there are two adjacent peaks in the dynamic change curve of light intensity. Since most passenger cars are equipped with two headlights, one on each side, there is a certain lateral offset between the two headlights and the rear-facing light sensor. When a vehicle approaches from directly behind, the light from the two headlights arrives at the sensor one after the other, or due to the angle difference, the light intensity received by the sensor shows two consecutive increases, first dominated by the left headlight and then by the right headlight, thus forming an imperfect bimodal structure in the curve. By detecting this bimodal structure, the system can effectively confirm that the target behind is a vehicle with two headlights and can distinguish it from single-headlight vehicles such as motorcycles and electric vehicles. This is helpful in assessing the type and size of the risk source. Peak plateau judgment refers to analyzing the trend of the curve after reaching the peak. If the dynamic change curve of light intensity maintains a flat range after reaching the peak, with the amplitude neither increasing nor decreasing significantly, it indicates that the vehicle behind has entered a following state with a relative speed close to zero, i.e., constant speed following. Conversely, if the curve drops rapidly after reaching its peak, it indicates that the vehicle behind is changing lanes to overtake or actively slowing down, thus reducing the risk of a rear-end collision. The peak platform assessment provides crucial information about the behavioral intentions of vehicles behind in risk assessment.

[0045] S103. Determine the risk level of the vehicle behind based on the described change characteristics.

[0046] Based on the changing characteristics identified in the above steps, the system can select at least one feature, or combine multiple features simultaneously, to determine the risk level of vehicles behind. The risk level can be determined through rule-based weighted scoring, or by using a machine learning model to fuse and infer the aforementioned features. This step, through multi-dimensional feature analysis, transforms the original light intensity curve into a quantified risk indicator, providing a reliable basis for subsequent graded early warning systems.

[0047] Before determining the risk level of following vehicles based on the aforementioned change characteristics, one possible implementation of this application further acquires vehicle status information for fusion evaluation with curve features. Vehicle status information refers to real-time data reflecting the vehicle's current driving environment and driving intentions, including but not limited to vehicle speed, vehicle acceleration, turn signal status, and weather and time information. Specifically, vehicle speed represents the vehicle's current speed, vehicle acceleration represents the degree to which the vehicle is accelerating or decelerating, turn signal status indicates whether the driver intends to change lanes or turn, and weather and time information provides environmental conditions such as rain, fog, and nighttime. This information is crucial for accurately assessing rear-end collision risks: for example, when the vehicle's speed is low or stationary, the relative speed from a vehicle approaching at the same speed is greater, resulting in a higher risk; when the vehicle is braking urgently, the risk of collision increases sharply if the vehicle behind does not decelerate; when the vehicle activates its turn signal to change lanes, the need for warnings to vehicles behind in the original lane can be appropriately reduced; and at night or in rainy or foggy weather, the system can appropriately increase its sensitivity to detect risks earlier.

[0048] After acquiring the vehicle's status information, this step inputs the features of the dynamic light intensity change curve and the vehicle's status information into a pre-trained machine learning model, which then outputs the risk level of the vehicles behind. A machine learning model is a data-driven computational model that learns the mapping relationship between input and output from a large number of samples, enabling it to predict new input data. The machine learning model used in this application is specifically a lightweight gradient boosting decision tree or a support vector machine. A lightweight gradient boosting decision tree is a gradient boosting algorithm based on decision tree ensembles, offering advantages such as fast training speed, low memory usage, and high prediction accuracy, making it suitable for real-time operation in automotive embedded environments. A support vector machine is a classification and regression model based on statistical learning theory, achieving nonlinear mapping by finding the optimal classification hyperplane, and is also suitable for real-time inference tasks.

[0049] To build this model, pre-training is required. The training process is as follows: First, real-vehicle data and simulation data are collected under different scenarios. These scenarios cover various approach modes of following vehicles, including normal following, high-speed rear-end collisions, interference from stationary high beams, nighttime driving, rain and fog, and different vehicle models. Each training sample consists of an input part and a label part. The input part includes the feature vector of the dynamic change curve of light intensity and the vehicle's state information vector. The label part includes the true risk level, the ground truth relative speed, and the ground truth relative distance corresponding to the scenario. The ground truth relative speed and relative distance can be simultaneously collected and calibrated using auxiliary equipment such as high-precision radar or lidar. The risk level is labeled by experts based on the actual collision risk. During training, the input feature vector is used as the model input, and the label is used as the expected output. The internal parameters of the model are continuously optimized through supervised learning algorithms so that the model's predicted output is as close as possible to the label for any given input. After training with a sufficient number of samples, the model learns the ability to infer the risk level, relative speed, and relative distance of following vehicles from the light intensity curve features and the vehicle's state.

[0050] In practical use, the system inputs the real-time extracted light intensity curve features and the real-time acquired vehicle status information into the trained model. After forward calculation, the model outputs the comprehensive risk level at the current moment, as well as relative speed and distance estimates. The comprehensive risk level can be an integer from 0 to 5, with higher values ​​indicating a greater degree of danger from approaching vehicles. This level will be directly used to trigger subsequent graded warning operations.

[0051] In one possible implementation, the output of the pre-trained machine learning model also includes relative speed estimates and distance estimates. The relative speed estimate refers to the approaching speed of a vehicle behind relative to the vehicle itself, measured in meters per second or kilometers per hour; a positive value indicates approaching, and a negative value indicates moving away. The distance estimate refers to the relative distance between the vehicle behind and the vehicle itself, measured in meters. These two estimates are not directly obtained from sensor measurements, but are inferred by the machine learning model based on the characteristics of the dynamic light intensity curve and the vehicle's state information.

[0052] Specifically, the determination of these two estimates relies on a light intensity model, the physical basis of which is the inverse square law of illuminance. According to this law, the illuminance E received by the rear-facing photosensitive sensor is directly proportional to the luminous intensity I of the rear vehicle's headlights and inversely proportional to the square of the relative distance d, i.e., E = I / d². Therefore, when the luminous intensity I of the rear vehicle's headlights is essentially constant, the signal amplitude output by the sensor is directly proportional to 1 / d², and the estimated value of the relative distance d can be derived by measuring the signal amplitude. Furthermore, by taking the first derivative of the signal amplitude over time, the rate of change of illuminance dE / dt can be obtained. This rate of change is related to the relative velocity v_rel as dE / dt ∝ (2·I·v_rel) / d³, meaning the rate of change of illuminance is related to the cube of the relative velocity and the distance. Therefore, the estimated value of the relative velocity can be derived from the rate of change of the signal amplitude.

[0053] In practical implementation, the system does not directly use the above analytical formula for calculation because the luminous intensity I may fluctuate due to factors such as vehicle model, headlight aging, and high / low beam status. Therefore, this application employs a machine learning model to learn the mapping relationship between the characteristics of the dynamic change curve of luminous intensity and relative speed and relative distance. During model training, a large amount of real vehicle data and simulation data are used as samples. Each sample contains multi-dimensional features of the dynamic change curve of luminous intensity, such as peak rate of change, standardized rise time, signal energy integral, etc., as well as the corresponding true values ​​of real relative speed and real relative distance. The true values ​​are synchronously collected and calibrated using auxiliary equipment such as high-precision radar or lidar. Through supervised learning, the model learns the mapping function from curve features to relative speed and relative distance. After training, during actual operation, the model can directly output the estimated values ​​of relative speed and distance by inputting the features of the current dynamic change curve of luminous intensity. This overcomes the difficulty of relying solely on physical formulas and requiring prior knowledge of the luminous intensity I, while fully utilizing the fitting ability of machine learning for nonlinear relationships, making the estimated values ​​more accurate and reliable.

[0054] S104. Execute the corresponding warning operation based on the risk level of the vehicle behind.

[0055] After obtaining the risk level of the vehicles behind, this step executes graded warning operations of different intensities and forms based on the level of risk and the expected collision time.

[0056] The risk levels are divided into five levels, with higher levels indicating a greater risk of rear-end collision from approaching vehicles. Specifically, Level 1 corresponds to a normal driving situation where the following vehicle is far away or traveling at a very low relative speed, resulting in an extremely low risk of collision; Level 2 corresponds to a situation where the following vehicle approaches at a relatively low speed, resulting in a low but noteworthy risk of collision; Level 3 corresponds to a situation where the following vehicle approaches at a medium to high speed, resulting in a moderate to high risk of collision, requiring the driver to be alerted; Level 4 corresponds to a situation where the following vehicle approaches at high speed, resulting in a high risk of collision, requiring a clear warning to the driver; and Level 5 corresponds to an emergency situation where the following vehicle approaches at extremely high speed or at very close range, resulting in an extremely high risk of collision, and a collision may be unavoidable. These five levels are progressive: Levels 1 and 2 are both in the low-risk range and share a preparatory warning operation, but their internal risk scores differ; the warning operation escalates from Level 3 onwards; Level 4 adds auditory cues; and Level 5 combines tactile cues and vehicle pre-control, thus forming a complete hierarchical response from imperceptible cues to multi-sensory, multi-modal warnings.

[0057] The estimated collision time is calculated by dividing the distance estimate by the relative speed estimate, in seconds. Its value directly reflects the urgency of the collision risk. This step presets three time thresholds: a first time threshold, a second time threshold, and a third time threshold. The first time threshold is greater than the second time threshold, and the second time threshold is greater than the third time threshold. For example, the first time threshold could be 7 seconds, the second time threshold could be 4 seconds, and the third time threshold could be 2 seconds, but in actual applications, these can be adjusted based on the vehicle model and calibration results.

[0058] When the risk level is Level 1 or Level 2, and the expected collision time exceeds the first time threshold, the system executes a preparatory warning operation. The preparatory warning operation is characterized by low disturbance and no interference, primarily used to enhance the driver's situational awareness without causing distraction. This operation includes presenting the driver with a visual cue with a first display effect when the system detects that the driver's gaze is directed towards the rearview mirror. Specifically, the first display effect visual cue can manifest as follows: when the driver actively observes the rearview mirror, a faint orange halo is superimposed on the virtual image of the mirror via an augmented reality head-up display. This halo does not emit an alarm or flash noticeably; it only appears when the driver is looking at it, thus subtly informing the driver that a vehicle is present behind them but does not pose an imminent danger. The driver's gaze being directed towards the rearview mirror can be detected by an in-vehicle driver monitoring camera or an infrared sensor installed near the rearview mirror. It should be noted that the preparatory warning operation is only triggered when the driver observes the rearview mirror. If the driver is not observing the mirror, no cue is issued to avoid unnecessary interference when the driver is focused on the road ahead.

[0059] When the risk level is Level 3 and the estimated collision time falls between the second and first time thresholds, the system executes a Level 1 warning. The Level 1 warning is a mild alert, requiring the driver sufficient time to observe and assess the situation behind. This operation includes presenting the driver with a visual cue with a second display effect. The second display effect differs significantly from the first display effect in display position, color, or brightness. Specifically, a yellow vehicle icon may be displayed in a fixed position on the instrument panel or head-up display. This icon remains continuously displayed without flashing or accompanied by an audible alert, aiming to gently alert the driver that a vehicle is approaching at medium to high speed from behind. The driver can actively check the rearview mirror to confirm the situation. Unlike the preparatory level warning, the Level 1 warning does not require detection of the driver's line of sight; instead, it actively presents the alert information within the driver's normal field of vision, ensuring the driver receives the alert even without checking the rearview mirror.

[0060] When the risk level is Level 4 and the estimated collision time falls between the third and second time thresholds, the system executes a Level 2 warning. The Level 2 warning is a clear alert requiring the driver to immediately pay attention to what's behind them. This action includes an audible cue with a first auditory effect and a visual cue with a third visual effect. The first auditory cue can be one or two medium-pitched beeps to attract the driver's attention without causing excessive panic. The third visual cue is more prominent than the second visual cue. Specifically, the yellow vehicle icon on the instrument panel or head-up display turns red and begins flashing at a higher frequency, while the LED strip integrated into the rearview mirror frame emits a red breathing flashing light effect. This light effect naturally guides the driver's gaze towards the rearview mirror, allowing the driver to quickly obtain information about what's behind them. The simultaneous presentation of visual and auditory cues in the Level 2 warning creates a sound-light linkage, significantly improving the urgency and recognizability of the warning.

[0061] When the risk level is Level 5 and the estimated collision time is less than the Level 3 time threshold, the system executes a Level 3 warning operation. The Level 3 warning operation is the highest level of emergency response, designed to fully attract the driver's attention and activate the passive safety system's pre-protection mechanism when a collision is unavoidable, reducing potential injury. This operation includes a visual cue (Level 4 display effect), an auditory cue (Level 2 auditory effect), a tactile cue, and pre-control vehicle actions. The visual cue (Level 4 display effect) is more intense than the Level 3 effect; in practice, the central control screen or full LCD instrument panel can instantly turn red and display a large collision warning symbol, creating a strong visual impact. The auditory cue (Level 2 auditory effect) is more urgent than the Level 1 effect; in practice, a unique, penetrating, continuous emergency alarm sound is emitted from the headrest speakers. Because the headrest speakers are close to the driver's ears, the sound is highly directional and not easily masked by other sounds inside the vehicle. The tactile cue is implemented by a vibration motor in the driver's seat emitting strong pulsed vibrations, directly stimulating the driver through tactile sensation.

[0062] The vehicle pre-control actions include, but are not limited to: controlling the seatbelt motor to pre-tension, eliminating the gap between the seatbelt and the occupant, and restraining the occupant's body in advance; adjusting the active headrests to the optimal protection position to reduce whiplash injuries to the neck in the event of a rear-end collision; closing all windows and sunroof to prevent occupants from being ejected or foreign objects from entering the vehicle during a collision; and controlling the exterior hazard lights to flash rapidly three to five times to send an emergency warning signal to following vehicles, reminding them to take braking or evasive action. The order and timing of the above vehicle pre-control actions can be flexibly adjusted according to the expected collision time and the vehicle's condition. For example, if the expected collision time is extremely short, only seatbelt pre-tensioning and headrest adjustment may be performed, while the window closing action may be omitted.

[0063] Through the aforementioned tiered early warning system, this application achieves a tiered response strategy that matches the risk level: the higher the risk level, the stronger the early warning operation, the more sensing channels, and the greater the urgency. Specifically, the first and second levels belong to the low-risk range, employing low-intrusion preparatory visual cues that only appear when the driver observes the rearview mirror; the third level belongs to the medium-risk range, upgrading to proactive visual cues that do not require detection of the driver's line of sight; the fourth level belongs to the high-risk range, adding auditory cues to the visual cues to create a sound-light linkage; and the fifth level belongs to the emergency risk range, further superimposing tactile cues and executing vehicle pre-control actions. This tiered design avoids driver fatigue and neglect caused by frequent alarms, while providing the strongest possible protection for occupants in truly dangerous situations.

[0064] In one possible implementation, this application can correlate the estimated relative speed and distance of following vehicles obtained by estimating the dynamic change curve of light intensity with a list of targets detected by millimeter-wave radar used in an existing blind spot monitoring system on the vehicle. Millimeter-wave radar is a sensor that uses electromagnetic waves in the millimeter-wave frequency band to detect the position and speed of targets. It is typically installed on both sides of the rear bumper of a vehicle to monitor vehicles approaching from the rear and sides.

[0065] This system matches its estimated relative speed and distance with the target detected by radar. If the radar also confirms the presence of the same target, the system can significantly increase the risk confidence level, and make more accurate decisions in subsequent warnings and control based on radar data, as radar has advantages in ranging and speed measurement accuracy. On the other hand, this system complements the radar: when the radar temporarily fails due to contamination such as being covered by mud or water, or due to strong electromagnetic interference, this system can temporarily take over the rear risk warning function, continuing to provide basic rear vehicle approach warnings, ensuring the continuity of the warning function and avoiding a lack of safety protection due to the failure of a single sensor.

[0066] In one possible implementation, this application embodiment further includes obtaining user feedback information on the warning operation and adjusting the user's personalized warning threshold within a safe range based on the feedback information. Feedback information refers to the driver's response to the warning operation issued by the system. For example, after the system issues a first-level warning operation, i.e., a visual cue, whether the driver observes the rearview mirror, takes evasive actions such as slowing down or changing lanes, or whether the driver frequently ignores a certain level of warning operation without taking any action.

[0067] The system can obtain the driver's gaze direction through the in-vehicle driver monitoring camera and acquire brake pedal signals, accelerator pedal signals, and turn signal status through the vehicle controller's local area network bus to determine whether the driver has responded effectively to the warning. If the system detects that a driver consistently ignores a specific level of warning, such as a preparatory warning, for an extended period of time—that is, neither checking the rearview mirror nor adjusting the vehicle speed—the system determines that the current warning threshold is too sensitive for that driver, resulting in unnecessary and frequent prompts.

[0068] In this situation, the system automatically fine-tunes the driver's personalized warning threshold within a safe range. For example, it might appropriately lower the first time threshold from 7 seconds to 6 seconds, or the second time threshold from 4 seconds to 3.5 seconds, ensuring that the corresponding level of warning is triggered only when a vehicle behind is closer or at a higher relative speed. The safe range refers to the threshold being neither too low, thus compromising the timeliness of the warning, nor too high, thus failing to provide an effective alert when danger arises. An adjustable upper and lower boundary is typically determined through calibration experiments. Through this online self-learning, the system can adapt to different drivers' reaction habits and risk preferences, reducing invalid alarms and improving the user experience.

[0069] In one possible implementation, this application embodiment further includes mirror surface dirt detection and functional failure handling. Specifically, when the signal amplitude of the rear-facing photosensor remains below a second amplitude threshold for a duration exceeding a first preset time, and the signal amplitude of the front-facing photosensor is within a preset normal range, the system generates a cleaning prompt. The second amplitude threshold is a preset lower limit; a value below this threshold indicates abnormally weak light intensity received by the rear-facing sensor. The first preset time is used to exclude brief changes in light or noise interference, and can be set to, for example, 5 seconds or 10 seconds. The forward-facing photosensor's signal amplitude being within the preset normal range indicates normal ambient lighting conditions in front of the vehicle, ruling out the possibility of a naturally low rear-facing signal due to a completely dark environment.

[0070] When the above conditions are met simultaneously, the system infers that dirt, dust, ice, snow, or mud have accumulated on the rear-view sensor's light-sensing window or the rearview mirror's surface, obstructing the normal incidence of light from following vehicles. The generated cleaning reminder message can be displayed on the instrument panel, announced via voice, or indicated by flashing icons, to inform the driver that the rearview mirror or the rear-view light sensor's light-sensing window needs cleaning to maintain the warning function's normal operation. Conversely, when the system detects that the rearview mirror's anti-glare function is off—for example, if the driver manually presses the anti-glare function's disable switch—or detects that the rear-view light sensor's light-sensing surface is obstructed by foreign objects or its light path is blocked (e.g., by attached decorations, sunshades, or physical obstacles in front of the sensor), the system determines that it cannot obtain a valid rear light signal. In this case, it automatically generates a warning message indicating that the warning function is unavailable. This message is also displayed on the instrument panel, audio system, or head-up display to inform the driver that the rear vehicle approach warning function is currently unavailable, reminding the driver to observe the road conditions behind using the traditional rearview mirror. In any of the above situations, the system will not execute the approach warning operation for vehicles behind, in order to avoid generating false or missed warnings due to invalid sensor signals.

[0071] In one possible implementation, this application embodiment also includes a degradation processing strategy for persistent signal anomalies. The system continuously monitors the signal quality of the rear and front photosensors. When a persistent signal anomaly is detected but the conditions for completely disabling the warning function have not yet been met—for example, the signal amplitude output by the rear photosensor is consistently high or low and fluctuates drastically, but the signal from the front photosensor is basically normal, or although the signal exists, the signal-to-noise ratio is too low to reliably extract the dynamic change curve of light intensity—the system determines that it is currently impossible to rely on optical signals for accurate risk assessment.

[0072] In this scenario, the system automatically degrades its operation, abandoning complex feature extraction and machine learning risk assessment using dynamic light intensity curves. Instead, it performs simple logical judgments based solely on the vehicle's speed (v_ego) and acceleration (a_ego). Vehicle speed (v_ego) refers to the vehicle's speed, and acceleration (a_ego) refers to its acceleration / deceleration state; both can be obtained in real-time from the vehicle controller's local area network bus or inertial measurement unit. The logical judgment rules in degraded mode can be pre-defined. For example, if the vehicle's acceleration (a_ego) is less than a preset emergency braking threshold (i.e., the vehicle is undergoing emergency braking with significant deceleration), and the vehicle speed (v_ego) is higher than a certain value, the system considers the vehicle to be in an emergency braking state. If following vehicles do not decelerate synchronously, the risk of a rear-end collision is extremely high. To provide a certain level of collision risk warning during sensor malfunctions, the system directly triggers an emergency warning operation, executing some or all of the third-level warning actions, such as issuing an emergency alarm, displaying a red warning on the dashboard, and tightening seat belts.

[0073] This degradation processing strategy ensures that even when optical sensor signals are abnormal, the system can still provide basic rear collision risk warnings using the vehicle's motion status information, avoiding the safety hazards caused by the complete loss of warning functions. When the signal returns to normal, the system automatically exits the degradation mode and resumes normal risk assessment and graded warning functions based on the dynamic change curve of light intensity.

[0074] The above are some specific implementations of the rear vehicle risk warning method provided in the embodiments of this application. Based on this, this application also provides a corresponding control system. The system provided in the embodiments of this application will be described below from the perspective of functional modularity. Figure 2 This is a structural diagram of a rear vehicle risk warning system provided in an embodiment of this application.

[0075] The system includes: The acquisition unit 110 is used to acquire the first signal of the rearward photosensitive sensor on the interior rearview mirror and the second signal of the forward photosensitive sensor on the interior rearview mirror. The first determining unit 111 is used to determine, based on the first signal and the second signal, the change characteristics used to characterize the change in the lights of the following vehicle; The second determining unit 112 is used to determine the risk level of the vehicle behind based on the changing characteristics; The execution unit 113 is used to perform corresponding early warning operations based on the risk level of the vehicle behind.

[0076] This application also provides corresponding devices and computer storage media for implementing the control scheme provided in this application.

[0077] The device includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code to enable the device to perform the rear vehicle risk warning method according to any embodiment of this application.

[0078] The computer storage medium stores code, and when the code is run, the device running the code implements the rear vehicle risk warning method according to any embodiment of this application.

[0079] This application also provides a corresponding vehicle, which includes the aforementioned equipment.

[0080] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0081] It should be understood that in this application, "at least one" refers to one or more items, and "more" refers to two or more items. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one" of a, b, or c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0082] It should be understood that the terms center, longitudinal, transverse, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0083] It should be noted that, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0084] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the statement "including a…" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0085] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for early warning of rear vehicle risks, characterized in that, include: Acquire the first signal from the rearward photosensitive sensor on the interior rearview mirror and the second signal from the forward photosensitive sensor on the interior rearview mirror; Based on the first signal and the second signal, determine the change characteristics used to characterize the changes in the lights of the vehicle behind; Based on the aforementioned changes, the risk level of the vehicles behind is determined; Based on the risk level of the vehicles behind, execute the corresponding warning operation.

2. The method according to claim 1, characterized in that, The step of determining the change features characterizing the changes in the headlights of vehicles behind based on the first signal and the second signal includes: Based on the second signal, ambient light compensation is performed on the first signal to obtain a third signal characterizing the contribution of the lights from vehicles behind. The dynamic change curve of the light intensity of the rear vehicle lights is extracted based on the third signal. Based on the dynamic change curve of light intensity, the change characteristics used to characterize the changes in the headlights of vehicles behind are determined.

3. The method according to claim 2, characterized in that, The extraction of the dynamic change curve of the light intensity of the rear vehicle lights based on the third signal includes: The third signal is subjected to bandpass filtering to obtain the filtered fourth signal; When the amplitude of the fourth signal is greater than or equal to the first amplitude threshold, the current moment is marked as the starting point of the optical event; Starting from the origin of the optical event, the amplitude of the fourth signal is recorded over time until the amplitude of the fourth signal is lower than the first amplitude threshold. The recorded time series data is then used as the dynamic change curve of the light intensity.

4. The method according to claim 2, characterized in that, The characteristics of the dynamic change curve of light intensity include at least one of the following: The signal peak change rate is the maximum value of the first derivative of the dynamic change curve of the light intensity, used to characterize the relative speed of the vehicle behind. Standardized rise time, which is the time required for the light intensity dynamic change curve to rise from the first percentage of the peak to the second percentage, is used to characterize the approach speed or initial distance of the following vehicle. The signal energy curve integral is the integral value of the dynamic change curve of light intensity on the time axis, which is used to characterize the light source intensity or proximity of the vehicle behind. Frequency domain modulation features, which are obtained by performing a short-time Fourier transform on the dynamic change curve of light intensity, include pulse features for identifying large vehicles or flicker features for identifying preset light source types. The double-peak detection result is used to determine whether there are two adjacent peaks in the curve, which is used to distinguish between vehicles with two lights and vehicles with one light. The peak platform judgment result is used to determine whether the curve maintains a flat range after reaching its peak, which is used to distinguish whether the following vehicles are following at a constant speed.

5. The method according to claim 1, characterized in that, Before determining the risk level of the vehicle behind based on the changing characteristics, the process also includes: Obtain vehicle status information, which includes at least one of vehicle speed, vehicle acceleration, turn signal status, and weather information; The process of determining the risk level of vehicles behind based on the aforementioned changes includes: The change features and the vehicle state information are input into a pre-trained machine learning model so that the machine learning model outputs the risk level of the vehicle behind. The pre-trained machine learning model is built based on a lightweight gradient boosting decision tree or support vector machine. It is obtained by collecting real vehicle data and simulation data in different scenarios, using the change features and the vehicle state information as input feature vectors, and using the corresponding risk level, relative speed estimate and distance estimate as labels for supervised learning training.

6. The method according to claim 5, characterized in that, The output of the pre-trained machine learning model also includes relative speed estimates and distance estimates; the step of executing corresponding warning operations based on the risk level of the vehicle behind includes: When the risk level is Level 1 or Level 2 and the estimated collision time is greater than the first time threshold, a preparatory warning operation is performed. The estimated collision time is the quotient of the distance estimate and the relative speed estimate. The preparatory warning operation includes a visual cue for the first display effect when the driver's line of sight is detected to be directed toward the interior rearview mirror. When the risk level is level three and the expected collision time is between the second time threshold and the first time threshold, a first-level warning operation is executed, which includes a visual cue with a second display effect. When the risk level is level four and the expected collision time is between the third time threshold and the second time threshold, a second-level warning operation is executed. The second-level warning operation includes an auditory prompt with a first auditory effect and a visual prompt with a third display effect. When the risk level is level 5 and the estimated collision time is less than the third time threshold, a level 3 warning operation is executed. The level 3 warning operation includes visual cues with a fourth display effect, auditory cues with a second auditory effect, tactile cues, and vehicle pre-control actions. The first time threshold is greater than the second time threshold, and the second time threshold is greater than the third time threshold.

7. The method according to claim 2, characterized in that, The step of performing ambient light compensation on the first signal based on the second signal to obtain a third signal characterizing the contribution of rear vehicle headlights includes: Subtracting the product of the second signal and the calibration coefficient from the first signal yields the third signal characterizing the contribution of the rear vehicle's headlights.

8. The method according to claim 1, characterized in that, Also includes: Obtain user feedback on the warning operation; Based on the feedback information, the user's personalized warning threshold is adjusted within a safe range.

9. The method according to claim 1, characterized in that, Also includes: When the signal amplitude of the rearward photosensitive sensor is continuously lower than the second amplitude threshold for a duration exceeding the first preset duration, and the signal amplitude of the forward photosensitive sensor is within the preset normal range, a cleaning reminder message is generated. The cleaning reminder message is used to remind the driver to clean the mirror surface of the rearview mirror or the light-sensing window of the rearward photosensitive sensor. When the anti-glare function of the rearview mirror is detected to be off, or when the photosensitive surface of the rear-facing photosensitive sensor is blocked by a foreign object or the light path is obstructed, a warning message indicating that the warning function is unavailable is generated.

10. A vehicle, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the rear vehicle risk warning method as described in any one of claims 1 to 9.

Images