An automatic driving safety control method and system based on environment perception

Abstract

The application is suitable for the field of automatic driving environment perception and safety control technology, and provides an automatic driving safety control method and system based on environment perception.The method comprises the following steps: analyzing historical environment perception data sets to screen out a plurality of reference samples matched with the current environment state characteristics, performing time sequence analysis on the reference samples to determine the visual recognition result stability index of the vehicle when passing through the light interference area in history and the polarization feature attenuation index of the corresponding building reflection structure; based on the multi-time sequence reference sample, the time variation trend of the visual recognition result stability index and the polarization feature attenuation index is analyzed to determine whether they present a preset coupling relationship.The application firstly considers the time sequence attenuation law of the polarization characteristics of the building reflection structure, compares the polarization feature attenuation indexes at different time sequences, quantifies the influence of polarization change on the visual recognition performance, and thus realizes the predictive correction of the weight before entering the light interference area.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of automatic driving environment perception and safety control, and particularly relates to an automatic driving safety control method and system based on environment perception. BACKGROUND

[0002] At present, automatic driving technology generally relies on multi-sensor fusion to realize comprehensive perception of the environment around the vehicle, among which the visual sensor plays a key role in target identification, lane keeping and obstacle detection. In order to ensure the reliability of multi-source information fusion, the vehicle-mounted perception system usually generates a visual fusion confidence weight value according to the real-time detected light intensity, image contrast, exposure information and recognition confidence, etc., to balance the influence weight of visual data and other sensor data such as radar and lidar. The existing algorithm mainly adjusts adaptively according to light intensity, shadow distribution and scene brightness change, and performs stably in most ordinary environments. However, in building environments with high reflectivity or strong polarization characteristics, such as glass curtain wall facades, mirror decorative outer walls and other areas, the visual sensor is easily disturbed by the polarization characteristics of reflected light, resulting in image information distortion and recognition performance degradation.

[0003] In actual application, when the vehicle drives to the same building reflection area for multiple times, the vehicle-mounted perception control system usually generates a visual fusion confidence weight value for multi-sensor fusion in advance according to the real-time detected light parameters before entering the light interference area. Since the external light conditions at this time are basically the same as when the vehicle first enters, the system usually considers that the environmental characteristics have not changed, thereby generating a visual fusion confidence weight value similar or identical to that when the vehicle first enters. However, the polarization characteristics of the building reflection structure will gradually deteriorate over time, for example, due to surface dust adhesion, coating aging or humidity change, causing changes in the polarization ratio and polarization angle of reflected light. The existing system does not have the ability to detect and model such polarization characteristic attenuation, so it cannot identify the correlation between the decline in visual recognition performance and the change in polarization characteristics, causing unbalanced weight distribution in the visual fusion stage, and thus reducing the stability and accuracy of target identification. SUMMARY

[0004] The present application aims to provide an automatic driving safety control method and system based on environment perception, which aims to solve the problems raised in the background art.

[0005] The present application is implemented as follows: an automatic driving safety control method based on environment perception, the method comprising:

[0006] When the vehicle is about to enter the light interference area formed by the specific building reflection structure in the high-frequency driving path, an initial visual fusion confidence weight value for multi-sensor fusion is extracted, and a historical environment perception data set of the vehicle on the high-frequency driving path is obtained;

[0007] The historical environment perception data set is analyzed to screen a plurality of reference samples matched with the current environment state feature, and the reference samples are time-series analyzed to determine a visual recognition result stability index of the vehicle when passing through the light interference area and a polarization feature attenuation index of the corresponding building reflection structure.

[0008] Based on the multi-time-series reference sample analysis of the visual recognition result stability index and the polarization feature attenuation index, whether the two present a preset coupling relationship is determined.

[0009] If it is determined that the preset coupling relationship exists, the current polarization feature attenuation index of the building reflection structure is obtained, which is differentially analyzed with the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value, a correction factor is generated, and the initial visual fusion confidence weight value is dynamically corrected by using the correction factor.

[0010] As a further limitation of the technical scheme of the embodiment of the application, the specific building reflection structure is a glass curtain wall facade with a high reflectivity coating or a polarization layer on the surface.

[0011] As a further limitation of the technical scheme of the embodiment of the application, the current environment state feature matching refers to the consistency between the reference sample and the light distribution, incident angle, environment brightness and reflection direction of the current environment of the vehicle being higher than a preset threshold.

[0012] As a further limitation of the technical scheme of the embodiment of the application, the historical environment perception data set includes multi-source perception data collected by the vehicle side and reflection feature data obtained by the building reflection structure side.

[0013] As a further limitation of the technical scheme of the embodiment of the application, the determination process of the visual recognition result stability index includes:

[0014] The multi-source perception data collected by the vehicle side is analyzed to determine the recognition confidence fluctuation amplitude, recognition result consistency and false detection rate of the vehicle visual recognition model in the reference sample when passing through the light interference area, and the recognition confidence fluctuation amplitude, recognition result consistency and false detection rate are integrated to generate a comprehensive evaluation value, which is used as the visual recognition result stability index.

[0015] As a further limitation of the technical scheme of the embodiment of the application, the calculation and acquisition process of the polarization feature attenuation index includes:

[0016] The reflection characteristic data obtained by the building reflection structure side is analyzed, the reflection light polarization rate and polarization angle information corresponding to the reference sample are extracted, and based on the reflection light polarization rate and polarization angle information relative to the initial visual fusion confidence weight value first formulated by the vehicle, the polarization rate decay amplitude and polarization angle offset are generated in turn, and the above results are comprehensively generated to generate the polarization characteristic decay index.

[0017] As a further limitation of the technical scheme of the embodiment of the application, the preset coupling relationship refers to:

[0018] When the change trend of the visual recognition result stability index gradually decreases, the change trend of the polarization characteristic decay index gradually increases, and the correlation coefficient between the change rate or change amplitude of the two is higher than the preset threshold, it is determined that there is a coupling relationship between the visual recognition performance decline and the polarization characteristic decay.

[0019] As a further limitation of the technical scheme of the embodiment of the application, if it is determined that there is a preset coupling relationship, the polarization characteristic decay index of the building reflection structure at present is obtained, and difference analysis is performed on the polarization characteristic decay index corresponding to the initial visual fusion confidence weight value first formulated by the vehicle, a correction factor is generated, and the initial visual fusion confidence weight value is dynamically corrected using the correction factor.

[0020] The polarization characteristic decay index of the building reflection structure at present is obtained, and the polarization characteristic decay index corresponding to the initial visual fusion confidence weight value first formulated by the vehicle is obtained based on the historical environment perception data set.

[0021] Difference analysis is performed on the above two polarization characteristic decay indexes to generate a correction factor, and the initial visual fusion confidence weight value is dynamically corrected using the correction factor.

[0022] The corrected visual fusion confidence weight value is obtained, and the vehicle passes through the current light interference area based on the corrected visual fusion confidence weight value.

[0023] As a further limitation of the technical scheme of the embodiment of the application, when the initial visual fusion confidence weight value is corrected, a preset correction function is used, and the correction function is:

[0024] ;

[0025] Wherein, The corrected visual fusion confidence weight value is denoted as W, The initial visual fusion confidence weight value is denoted as W0, The upper limit of the visual fusion confidence weight value is denoted as Wmax, The polarization characteristic decay index of the building reflection structure at present is denoted as P, refers to the polarization feature attenuation index corresponding to when the vehicle first formulates the initial visual fusion confidence weight value, refers to the correction factor, refers to the control correction amplitude coefficient.

[0026] An automatic driving safety control system based on environment perception, the system comprises:

[0027] A weight extraction module is configured to extract an initial visual fusion confidence weight value for multi-sensor fusion when the vehicle is about to enter a light interference area formed by a specific building reflection structure in a high-frequency driving path, and obtain a historical environment perception data set of the vehicle on the high-frequency driving path.

[0028] A sample analysis module is configured to analyze the historical environment perception data set to filter out a plurality of reference samples matched with the current environment state feature, and perform time sequence analysis on the reference samples to determine a visual recognition result stability index of the vehicle when passing through the light interference area and a polarization feature attenuation index of the corresponding building reflection structure.

[0029] A coupling determination module is configured to analyze the time variation trend of the visual recognition result stability index and the polarization feature attenuation index based on multi-time sequence reference samples to determine whether they present a preset coupling relationship.

[0030] A dynamic correction module is configured to, if it is determined that the preset coupling relationship exists, obtain the current polarization feature attenuation index of the building reflection structure, perform difference analysis on the polarization feature attenuation index corresponding to when the vehicle first formulates the initial visual fusion confidence weight value, generate a correction factor, and perform dynamic correction on the initial visual fusion confidence weight value by using the correction factor.

[0031] Compared with the prior art, the present application has the following beneficial effects:

[0032] By introducing the coupling analysis mechanism between the polarization feature attenuation index and the visual recognition result stability index, the present application realizes dynamic adaptive correction of the visual fusion confidence weight value. Unlike the prior art which only sets the weight based on the light intensity or brightness parameter, the present application first considers the time sequence attenuation law of the polarization characteristics of the building reflection structure, compares the polarization feature attenuation indexes at different time sequences, quantifies the influence of polarization change on the visual recognition performance, and thus realizes predictive correction of the weight before entering the light interference area.

[0033] The method not only can accurately identify the recognition performance decline caused by the deterioration of the polarization characteristics, but also can realize dynamic balance of the visual channel weight in multi-sensor fusion, greatly improves the perception stability and driving safety of the vehicle in the high reflection and complex light environment, and has significant engineering application value and innovative significance. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 A flowchart of the method provided by the embodiment of the present application is shown.

[0035] Figure 2 A flowchart of the method provided by the embodiment of the present application is shown.

[0036] Figure 3 An application architecture diagram of the system provided by the embodiment of the present application is shown. DETAILED DESCRIPTION

[0037] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0038] Figure 1 A flowchart of the method provided by the embodiment of the present application is shown.

[0039] Specifically, an automatic driving safety control method based on environment perception, which specifically comprises the following steps:

[0040] Step S100, when the vehicle is about to enter a light interference area formed by a specific building reflection structure in a high-frequency driving path, an initial visual fusion confidence weight value for multi-sensor fusion is extracted, and a historical environment perception data set of the vehicle on the high-frequency driving path is obtained. The specific building reflection structure is a glass curtain wall facade with a high reflectivity coating or a polarizing layer on the surface.

[0041] The historical environment perception data set includes multi-source perception data collected by the vehicle and reflection feature data obtained by the building reflection structure.

[0042] In the embodiment of the present application, the vehicle refers to an intelligent automobile with automatic driving function and integrated vehicle visual recognition model. The vehicle can obtain environmental information through a visual perception system, a laser radar, a millimeter wave radar, an infrared sensor and other types of sensors, and realize functions such as automatic navigation, obstacle recognition, path planning and environment understanding based on multi-source perception results. The visual recognition model of the vehicle can perceive and recognize environmental factors such as external light, reflection and occlusion, thereby providing basic data for the generation of visual fusion confidence weight values.

[0043] The high-frequency driving path refers to a route that the vehicle repeatedly drives many times in daily operation, such as a city commuting trunk road, a fixed regional connection route, or a logistics distribution route, etc. In such a path, the long-term operation of the vehicle can accumulate a relatively rich historical environmental perception data, so that the system can analyze the trend and adapt to the lighting conditions, building reflection characteristics and traffic state of the same area, thereby improving the environmental adaptability of autonomous driving.

[0044] The specific building reflection structure is a building facade located on one side of the road or in the adjacent area of the driving path, which has a glass curtain wall structure with a high reflectivity coating or a polarization layer on its surface. Such structures are usually found on the outer facades of city high-rise buildings, and the reflected light not only has a high brightness, but also has obvious polarization characteristics. When sunlight or other strong light sources shine on such building surfaces, a directional and polarized reflected light distribution is formed, which may form local overexposure, reflected glare or visual offset in the perception range of the vehicle camera system, thereby affecting the visual recognition performance of the autonomous driving system.

[0045] The light interference area refers to a specific spatial area formed by the above-mentioned building reflection structure, in which the light distribution is strongly affected by reflection, and the light intensity, incident angle and polarization state are significantly different from the surrounding environment. When the vehicle enters this area, the light signal received by its visual sensor contains a strong reflected interference component, which can easily lead to a decrease in the stability of the visual recognition result or a visual fusion misjudgment.

[0046] The visual fusion confidence weight value is a weighting coefficient set by the vehicle perception system in the multi-sensor fusion process to represent the reliability of the visual recognition result. This weight value is calculated by the vehicle-mounted perception control system based on environmental light intensity, image contrast, recognition confidence distribution, sensor consistency and other factors, and is a relatively mature weight adjustment method in existing autonomous driving technology.

[0047] In a popular understanding, this weight value is equivalent to a quantitative indicator of the degree to which the system trusts the camera. When the environmental light is stable, the reflection interference is small, and the image quality is clear, the system will give a high confidence weight to the visual recognition, so that the visual sensor will dominate in the fusion decision; when the light is abnormal, the reflection is strong, or the recognition fluctuation is large, the system will reduce the weight of the visual channel and enhance the dependence on radar, infrared and other sensor data, so as to ensure the stability and safety of the overall perception result.

[0048] The historical environment perception data set includes multi-source perception data collected by a vehicle side and reflection feature data acquired by a building reflection structure side. The multi-source perception data collected by the vehicle side includes vehicle camera images, radar echo signals, ambient light intensity parameters, sensor synchronization timestamps, and recognition confidence, etc., which are used to reflect the perception performance of the vehicle in the historical driving process. The reflection feature data acquired by the building reflection structure side includes reflection light polarization rate, polarization angle, reflection intensity distribution, and light incidence angle, etc. parameters, which are obtained by an external monitoring device based on environment perception or by a vehicle perception system side, and are used to reflect the physical property changes of the building surface reflection light.

[0049] The core research point of the present application is that the existing automatic driving system mainly sets the weight according to the static light features such as ambient light intensity, image contrast, and recognition confidence when generating the visual fusion confidence weight value, without considering the dynamic attenuation effect of the polarization characteristics of the building reflection structure. When the vehicle first passes through the light interference area formed by the specific building reflection structure, the system can generate an adaptive visual fusion confidence weight value according to the light and reflection characteristics at that time. However, when the vehicle subsequently drives to the same area multiple times, the vehicle-mounted perception control system will generally generate a visual fusion confidence weight value for multi-sensor fusion according to the real-time detected light parameters before entering the light interference area. Since the external light conditions at this time are basically the same as when the vehicle first enters, the system generally considers that the environmental characteristics have not changed, thereby generating a visual fusion confidence weight value that is similar or the same as when the vehicle first enters.

[0050] In fact, the polarization characteristics of the building reflection structure will gradually decay over time and with the accumulation of surface attachments, and the polarization rate and polarization angle of the reflected light will shift, so that the stability of the visual recognition result will obviously decrease when the vehicle uses the uncorrected weight value. The existing technology does not consider the dynamic degradation of the polarization characteristics, and cannot extract the influence rule of the polarization change on the visual fusion performance from the historical environment perception data, resulting in that the system cannot realize reasonable weight adjustment under the condition that the light conditions are seemingly consistent.

[0051] Further, the automatic driving safety control method based on environment perception further comprises the following steps:

[0052] Step S200, analyze the historical environment perception data set to screen out a plurality of reference samples matched with the current environmental state characteristics, and perform time sequence analysis on the reference samples to determine the visual recognition result stability index of the vehicle when passing through the light interference area and the polarization feature decay index of the corresponding building reflection structure.

[0053] The current environment state feature matching refers to a case that consistency between the reference sample and the light distribution, the incident angle, the environment brightness and the reflection direction of the current environment of the vehicle is higher than a preset threshold.

[0054] The determination process of the visual recognition result stability index includes: analyzing the multi-source perception data collected by the vehicle side, determining the recognition confidence fluctuation amplitude, the recognition result consistency and the false detection rate of the vehicle visual recognition model in the reference sample when passing through the light interference area, and generating a comprehensive evaluation value by comprehensively combining the recognition confidence fluctuation amplitude, the recognition result consistency and the false detection rate, and taking the comprehensive evaluation value as the visual recognition result stability index.

[0055] The calculation and acquisition process of the polarization feature attenuation index includes: analyzing the reflection feature data acquired by the building reflection structure side, extracting the reflection light polarization rate and polarization angle information corresponding to the reference sample, and based on the reflection light polarization rate and polarization angle information relative to the initial visual fusion confidence weight value when the vehicle is first formulated, generating the polarization rate attenuation amplitude and the polarization angle offset in turn, and generating the polarization feature attenuation index by comprehensively combining the above results.

[0056] In the embodiment of the application, in order to accurately evaluate the change trend of the visual recognition stability and the environmental polarization characteristics of the vehicle in the light interference area, it is necessary to analyze the historical environment perception data set to screen out a plurality of reference samples highly matched with the current environment state features. The reason for this screening is that there may be slight differences in external environment parameters (such as solar elevation angle, cloud distribution, aerosol concentration, etc.) when the vehicle passes through the same road section at different times, and these differences will significantly affect the polarization characteristics of the reflected light and the visual recognition performance. If strict screening is not performed, the analysis of the visual recognition result stability index and the polarization feature attenuation index will be deviated, thereby affecting the accuracy of the correction factor.

[0057] Therefore, the application constructs an environment state feature matching model to perform multi-dimensional similarity calculation on each sample in the historical environment perception data set, and screens out a sample set highly consistent with the current environment state of the vehicle. The current environment state features not only include basic parameters such as light distribution, incident angle, environment brightness and reflection direction, but also can further include environmental color temperature, weather type, solar radiation intensity, geographical azimuth angle corresponding to the driving time, suspended particle concentration in the air, etc. By introducing these additional feature parameters, the high comparability of the selected samples in the light environment and the reflection condition can be effectively ensured, thereby ensuring the stability and reliability of the time sequence analysis result.

[0058] In the process of determining the visual recognition result stability index, the system calls the vehicle multi-source perception result recorded in the historical running data for each reference sample meeting the screening condition, and restores and analyzes the recognition performance of the vehicle visual recognition model when passing through the light interference area. Specifically, when the vehicle passes through the light interference area, the visual fusion confidence weight value is usually generated according to the current environment light state before entering the area, and the actual application effect of the weight value can be inversely evaluated through the historical recognition data. The system uses the multi-source sensor data fusion algorithm to analyze the confidence fluctuation curve of the visual recognition output, combines the consistency test and false rejection rate calculation of the time sequence recognition result, constructs a comprehensive evaluation model, generates a stability comprehensive evaluation value in a mathematical weighted manner, and takes the value as the visual recognition result stability index. The processing process can be realized by a deep belief network (DBN) model to ensure that the index reflects the true recognition stability performance of the vehicle in the area.

[0059] The polarization feature attenuation index is used to represent the degree of attenuation of the polarization characteristics of the building reflection structure over time, which is one of the important innovations of the present application. The calculation process of the index includes: first, analyzing the reflection feature data obtained from the side of the building reflection structure, which can be obtained in real time by a polarization imaging device, a spectral reflection sensor or a polarization camera installed on the roadside or the outside of the vehicle, for recording the polarization ratio, polarization angle and spectral distribution of the reflected light of the building surface, etc. information of the incident light; then extract the reflected light polarization ratio, polarization angle and polarization spectral intensity distribution information corresponding to the reference sample, and compare these data with the reflection characteristics corresponding to the initial visual fusion confidence weight value when the vehicle first formulates it, and calculate the polarization ratio attenuation amplitude and polarization angle offset.

[0060] In addition, in order to improve the stability and representativeness of the index, the present application can further introduce the reflection spectral width change, the polarization light reflectivity ratio and the time difference parameters of the reflected light intensity as auxiliary dimensions for weighted fusion to generate the polarization feature attenuation index comprehensively.

[0061] The unique feature of the polarization feature attenuation index is that it not only reflects the physical degradation of the building reflection surface (such as surface dust, coating aging or micro scratches), but also quantifies the change trend of the reflection property over time through the dynamic characteristics of the polarization light, so that the system can identify the subtle deterioration of the polarization characteristics under the condition that the macroscopic light condition does not change. The traditional automatic driving vision system only estimates the reflection interference based on the brightness or contrast parameters, and cannot identify the change of this "polarization dimension". Therefore, by introducing the polarization feature attenuation index, the system can dynamically adjust the visual fusion confidence weight under the condition that the environment light is stable, realize the adaptive compensation of the polarization characteristic change, and significantly improve the visual fusion accuracy and safety in complex urban reflection environment.

[0062] Further, the environment perception based automatic driving safety control method further comprises the following steps:

[0063] In step S300, the time variation trend of the visual recognition result stability index and the polarization feature attenuation index is analyzed based on the multi-time sequence reference samples to determine whether the two present a preset coupling relationship.

[0064] The preset coupling relationship refers to that when the variation trend of the visual recognition result stability index gradually decreases, the variation trend of the polarization feature attenuation index gradually increases, and the correlation coefficient between the variation rate or variation amplitude of the two is higher than a preset threshold, it is determined that there is a coupling relationship between the visual recognition performance decline and the polarization feature attenuation.

[0065] In the embodiment of the application, the core of step S300 is to identify whether there is a significant coupling relationship between the visual recognition result stability index and the polarization feature attenuation index through comparative analysis of the multi-time sequence reference samples. Since the external environment lighting conditions, incident angles, driving speeds and sensor working states remain basically unchanged when the vehicle passes through the same building reflection area multiple times, when the stability of the visual recognition result continuously decreases, the main difference factor is usually derived from the change of the polarization characteristic of the building reflection structure. However, in order to avoid misjudgment caused by accidental noise or other unknown factors, the correlation between the two needs to be quantitatively identified and verified through a preset coupling relationship model.

[0066] The preset coupling relationship is an empirical model established based on a large amount of experimental data and deep learning technology. Through statistical analysis of the visual recognition performance under different polarization feature attenuation conditions in the laboratory and actual road environment, a functional mapping relationship between the polarization feature attenuation index and the visual recognition result stability index is constructed. The system uses a multi-time sequence data analysis model (such as a time sequence convolution network TCN or a long short-term memory network LSTM) to perform trend fitting and correlation calculation on the index sequence of the reference samples. When the visual recognition result stability index presents a gradual downward trend, the polarization feature attenuation index presents a gradual upward trend, and the correlation coefficient between the variation rate or amplitude of the two is higher than a preset threshold, it is determined that there is a coupling relationship between the visual recognition performance decline and the polarization feature attenuation.

[0067] Through the establishment and identification of the preset coupling relationship, the system can accurately locate the real reason for the degradation of visual recognition under the condition that the lighting condition is stable, thereby providing a scientific basis for the correction of the subsequent visual fusion confidence weight value.

[0068] Further, the environment perception based automatic driving safety control method further comprises the following steps:

[0069] Step S400, if it is determined that the preset coupling relationship exists, the current polarization feature attenuation index of the building reflection structure is obtained, difference analysis is performed on the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value, a correction factor is generated, and the initial visual fusion confidence weight value is dynamically corrected by using the correction factor.

[0070] Specifically, Figure 2 A flowchart for correcting the initial visual fusion confidence weight value is shown.

[0071] If it is determined that the preset coupling relationship exists, the current polarization feature attenuation index of the building reflection structure is obtained, difference analysis is performed on the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value, a correction factor is generated, and the initial visual fusion confidence weight value is dynamically corrected by using the correction factor. Specifically, the following steps are included:

[0072] Step S401, the current polarization feature attenuation index of the building reflection structure is obtained, and the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value is obtained based on a historical environment perception data set;

[0073] Step S402, difference analysis is performed on the above two polarization feature attenuation indexes to generate a correction factor, and the initial visual fusion confidence weight value is dynamically corrected by using the correction factor;

[0074] Step S403, the corrected visual fusion confidence weight value is obtained, and the vehicle passes through the current light interference area based on the corrected visual fusion confidence weight value.

[0075] In the correction of the initial visual fusion confidence weight value, a preset correction function is used, and the correction function is:

[0076] ;

[0077] Among them, refers to the corrected visual fusion confidence weight value, refers to the initial visual fusion confidence weight value, refers to the upper limit of the visual fusion confidence weight value, refers to the current polarization feature attenuation index of the building reflection structure, refers to the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value, refers to the correction factor, refers to the control correction amplitude coefficient.

[0078] In the embodiment of the present application, the core of step S400 is that after identifying that the visual recognition result stability index and the polarization characteristic attenuation index exist a preset coupling relationship, the initial visual fusion confidence weight value is dynamically corrected to adapt to the change trend of the polarization characteristic of the building reflection structure.

[0079] Since the polarization characteristic is a physical property that gradually attenuates over time, when the building surface coating layer or polarization layer is affected by environmental aging, dust deposition, humidity erosion and other factors, its polarization ratio and polarization angle will change, thereby the polarization characteristic of the reflected light is no longer consistent with the reference characteristic when the vehicle first formulates the initial visual fusion confidence weight value. If the old weight value is still used before entering the light interference area under the current condition, it will lead to that the confidence estimation of the visual recognition model to the reflected target is too high, thereby affecting the accuracy of the fusion decision.

[0080] Therefore, the present application proposes the idea of weight correction based on two groups of polarization characteristic attenuation indexes with different time sequences. By comparing the current polarization characteristic attenuation index of the building reflection structure with the corresponding polarization characteristic attenuation index when the vehicle first formulates the initial visual fusion confidence weight value, the system can accurately quantify the change degree of the polarization characteristic, thereby generating a correction factor reflecting the influence intensity of polarization attenuation. The significance of this correction method is that:

[0081] On the one hand, it is based on the adaptive weight adjustment of physical characteristic change, rather than relying on empirical threshold, so it has higher scientificity and interpretability;

[0082] On the other hand, by using the comparison relationship between the historical reference state and the current state, the interference of noise, instantaneous light fluctuation and other factors can be effectively avoided, so that the weight correction is more stable and reliable.

[0083] In the present embodiment, the specific execution process of step S403 is as follows:

[0084] After the correction module calculates the corrected visual fusion confidence weight value, it is input into the multi-sensor fusion algorithm of the vehicle perception control system in real time, which is used to replace the original fixed weight parameter. When the system fuses multi-source perception data from cameras, radars, lidar and the like, the credibility proportion of the visual recognition result in the overall fusion model is redistributed according to the corrected weight value, so that the vehicle can more accurately identify the front obstacles and road structures when passing through the light interference area, thereby improving the safety and robustness of the automatic driving decision.

[0085] The technical solutions of the present application are described below through a specific implementation example.

[0086] When the vehicle first passes through the glass curtain wall area on the high-frequency driving path, the vehicle-mounted perception control system calculates the visual fusion confidence weight value as 0.80 according to the light intensity, image contrast and sensor consistency at that time, and records the polarization characteristic attenuation index of the building reflection structure as 0.10. The data is stored as an initial reference in the historical environment perception data set.

[0087] Three weeks later, the vehicle drives to the same area again. Since the light distribution, incident angle and environmental brightness are basically the same as the first time, the system automatically calls the same initial weight value of 0.80 according to the current environmental state. However, in the historical data comparison, it is found that the polarization characteristic attenuation index has increased from 0.10 to 0.25, and the visual recognition result stability index has decreased from 0.92 to 0.78. The system judges that the change direction of the two is opposite, and the correlation coefficient reaches 0.82, which exceeds the preset threshold value of 0.70, confirming that there is a preset coupling relationship.

[0088] Then enter the correction stage. The system extracts the difference of 0.15 between the current polarization characteristic attenuation index 0.25 and the initial index 0.10, and calculates the polarization attenuation ratio as 1.5. According to the correction coefficient 0.1, the weight correction ratio is 1 minus 0.1 times 1.5, the result is 0.85. The system multiplies the initial weight value 0.80 by 0.85 to obtain the corrected visual fusion confidence weight value as 0.68, and updates it to the multi-sensor fusion module in real time.

[0089] When the vehicle enters the light interference area with the corrected weight, the system reduces the weight proportion of visual recognition in the fusion model, and improves the participation of radar and lidar. The test results show that the visual recognition result stability has increased from 0.74 to 0.88, and the false detection rate has decreased from 12% to 6%, and the vehicle can smoothly pass through the high reflection area.

[0090] This method compares two groups of polarization characteristic attenuation indexes with different time sequences, quantifies the influence of polarization change, and realizes the adaptive correction of the visual channel weight. Its advantages are intuitive calculation, rapid response and avoidance of noise misjudgment. Through this correction mechanism, the vehicle can maintain recognition stability when the polarization characteristics deteriorate under the same light conditions, significantly improving the safety, reliability and practical value of autonomous driving in complex reflection environments.

[0091] In summary, the overall beneficial effects of the present application are: by introducing the polarization characteristic attenuation index to dynamically correct the visual fusion confidence weight value, the adaptive perception and response of the polarization characteristic change of the autonomous driving system are first realized. Compared with the prior art, the present scheme can identify the visual recognition performance degradation problem caused by the deterioration of polarization characteristics under the condition that the light conditions are basically unchanged, thereby effectively solving the core problem proposed in step S100 that the existing system cannot identify the polarization change and correct the weight.

[0092] The technical solution can be widely applied to automatic driving scenes in urban high-reflection environments, such as glass curtain wall dense road sections in urban core areas, airport entrances and exits, commercial complex roads, and the like. The core algorithm module can be embedded in an existing vehicle perception control system, without the need to increase additional hardware costs, so as to improve the recognition stability and safety of the vehicle in a complex light interference environment, and has a significant promoting effect on the practicality and reliable operation of the automatic driving system.

[0093] Further, Figure 3 An application architecture diagram of the system provided by the embodiment of the application is shown.

[0094] In another preferred embodiment provided by the application, an automatic driving safety control system based on environmental perception comprises:

[0095] The weight extraction module 100 is configured to extract an initial visual fusion confidence weight value for multi-sensor fusion when the vehicle is about to enter a light interference area formed by a specific building reflection structure in a high-frequency driving path, and obtain a historical environmental perception data set of the vehicle on the high-frequency driving path.

[0096] Further, the automatic driving safety control system based on environmental perception further comprises:

[0097] The sample analysis module 200 is configured to analyze the historical environmental perception data set to screen a plurality of reference samples matched with the current environmental state features, and perform time sequence analysis on the reference samples to determine a visual recognition result stability index of the vehicle when passing through the light interference area and a polarization feature attenuation index of the corresponding building reflection structure.

[0098] Further, the automatic driving safety control system based on environmental perception further comprises:

[0099] The coupling determination module 300 is configured to analyze the time variation trend of the visual recognition result stability index and the polarization feature attenuation index based on the multi-time sequence reference samples, to determine whether the two present a preset coupling relationship.

[0100] Further, the automatic driving safety control system based on environmental perception further comprises:

[0101] The dynamic correction module 400 is configured to, if it is determined that the preset coupling relationship exists, obtain the current polarization feature attenuation index of the building reflection structure, perform difference analysis on the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value, generate a correction factor, and utilize the correction factor to dynamically correct the initial visual fusion confidence weight value.

[0102] It should be understood that, although the steps in the flowcharts of the embodiments of the present application are shown in a certain order according to the arrows, the steps are not necessarily executed in the order of the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in order, and the steps can be executed in other orders. Moreover, at least some of the steps in the embodiments can include a plurality of sub-steps or a plurality of stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the sub-steps or stages is not necessarily sequential, but can be round-robin or alternately executed with at least some of the other steps or sub-steps or stages of the other steps.

[0103] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a non-volatile computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of the methods. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0104] The technical features of the above-mentioned embodiments can be combined in any way. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0105] The above embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but should not be understood as a limitation on the patent scope of the present application. It should be noted that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

[0106] The above merely describes the preferred embodiments of the present application and should not be used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. An environment perception-based automatic driving safety control method, characterized by, The method comprises: When the vehicle is about to enter the light interference area formed by the specific building reflection structure in the high-frequency driving path, the initial visual fusion confidence weight value for multi-sensor fusion is extracted, and the historical environment perception data set of the vehicle on the high-frequency driving path is obtained; The historical environment perception data set is analyzed to screen out a plurality of reference samples matched with the current environment state characteristics, and the reference samples are time-series analyzed to determine the visual recognition result stability index of the vehicle when passing through the light interference area and the polarization feature attenuation index of the corresponding building reflection structure; The time variation trend of the visual recognition result stability index and the polarization feature attenuation index is analyzed based on the multi-time-series reference samples to determine whether they present a preset coupling relationship; If it is determined that the preset coupling relationship exists, the current polarization feature attenuation index of the building reflection structure is obtained, which is differentially analyzed with the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value, a correction factor is generated, and the initial visual fusion confidence weight value is dynamically corrected by using the correction factor; The preset coupling relationship refers to: when the change trend of the visual recognition result stability index gradually decreases, and the change trend of the polarization feature attenuation index gradually increases, and the correlation coefficient between the change rate or change amplitude of the two is higher than a preset threshold, it is determined that there is a coupling relationship between the visual recognition performance decline and the polarization feature attenuation; This step specifically comprises: The current polarization feature attenuation index of the building reflection structure is obtained, and the polarization feature attenuation index corresponding to the initial visual fusion confidence weight value when the vehicle first formulates the initial visual fusion confidence weight value is obtained based on the historical environment perception data set; The two polarization feature attenuation indexes are differentially analyzed to generate a correction factor, and the initial visual fusion confidence weight value is dynamically corrected by using the correction factor; The corrected visual fusion confidence weight value is obtained, so that the vehicle passes through the current light interference area based on the corrected visual fusion confidence weight value; When correcting the initial visual fusion confidence weight value, a preset correction function is used, and the correction function is: ； wherein, refers to the modified visual fusion confidence weight value, refers to the initial visual fusion confidence weight value, refers to the upper limit of the visual fusion confidence weight value, refers to the current polarization feature decay index of the building reflective structure, refers to the polarization feature decay index corresponding to the first time the vehicle formulates the initial visual fusion confidence weight value, refers to the correction factor, refers to the control correction amplitude coefficient. 2.The environment perception based automatic driving safety control method of claim 1, wherein, The specific building reflection structure is a glass curtain wall facade with a high reflectivity coating or a polarization layer on the surface. 3.The environment perception based automatic driving safety control method of claim 1, wherein, The current environment state characteristic matching refers to the consistency between the reference sample and the light distribution, incident angle, environment brightness and reflection direction of the current environment of the vehicle being higher than a preset threshold. 4.The environment perception based automatic driving safety control method of claim 1, wherein, The historical environment perception data set comprises multi-source perception data collected by the vehicle side and reflection feature data obtained by the building reflection structure side. 5.The environment perception based automatic driving safety control method of claim 4, wherein, The determination process of the visual recognition result stability index comprises: The multi-source perception data collected by the vehicle side is analyzed to determine the recognition confidence fluctuation amplitude, recognition result consistency and false detection rate of the vehicle visual recognition model in the reference sample when passing through the light interference area, and the recognition confidence fluctuation amplitude, recognition result consistency and false detection rate are combined to generate a comprehensive evaluation value, which is used as the visual recognition result stability index. 6.The environment perception based automatic driving safety control method of claim 4, wherein, The calculation and acquisition process of the polarization feature attenuation index comprises: The reflection characteristic data obtained by the building reflection structure side is analyzed, the reference sample corresponding reflection light polarization rate and polarization angle information are extracted, and based on the reflection light polarization rate and polarization angle information relative to the first time the vehicle formulates the initial visual fusion confidence weight value, polarization rate attenuation amplitude and polarization angle offset are generated in turn, and the above results are integrated to generate a polarization characteristic attenuation index.

7. A system for performing the environment-aware based automatic driving safety control method according to any one of claims 1-6, characterized in that, The system comprises: A weight extraction module is configured to extract an initial visual fusion confidence weight value for multi-sensor fusion when the vehicle is about to enter a light interference area formed by a specific building reflection structure in a high-frequency driving path, and obtain a historical environment perception data set of the vehicle on the high-frequency driving path; A sample analysis module is configured to analyze the historical environment perception data set to screen out a plurality of reference samples matching the current environment state characteristics, and to perform time series analysis on the reference samples to determine the visual recognition result stability index of the vehicle when passing through the light interference area and the polarization characteristic attenuation index of the corresponding building reflection structure; A coupling determination module is configured to analyze the time variation trend of the visual recognition result stability index and the polarization characteristic attenuation index based on multi-time series reference samples to determine whether they present a preset coupling relationship; A dynamic correction module is configured to, if it is determined that the preset coupling relationship exists, obtain the current polarization characteristic attenuation index of the building reflection structure, perform difference analysis on the polarization characteristic attenuation index corresponding to the first time the vehicle formulates the initial visual fusion confidence weight value, generate a correction factor, and dynamically correct the initial visual fusion confidence weight value using the correction factor.

Images