Lateral backward visual safety early warning method based on deep learning

Abstract

The invention discloses a side backward visual safety early warning method based on deep learning, and particularly relates to the field of safety early warning, comprising time region division, side backward data acquisition, side backward data analysis, comprehensive analysis and safety early warning. According to the method, comprehensive and real-time monitoring of the side backward safety of the target vehicle is achieved through the deep learning technology, potential safety risks can be found and early warned in time through refined time region division, multi-dimensional data collection and analysis, establishment of a comprehensive evaluation model and triggering of an early warning mechanism, and the safety of the target vehicle is improved. The method not only improves the running safety of the vehicle, but also reduces the accident rate caused by the lateral backward safety problem, provides powerful technical support for the safe running of the vehicle, and has important practical application value and popularization significance.

Description

Technical Field

[0001] The present invention relates to the field of safety warning technology, and more specifically, to a side and rear visual safety warning method based on deep learning. Background Art

[0002] With the continuous development of intelligent transportation systems, vehicle safety warning technology has become one of the important means to ensure road driving safety. Especially in the field of vehicle side and rear safety monitoring, by real-time monitoring of the traffic environment behind the vehicle, it can effectively prevent traffic accidents caused by blind spots or emergencies.

[0003] Most of the existing vehicle side rear safety warning technologies rely on sensor networks and algorithm models. Its operation process usually includes data collection, target detection, risk assessment and early warning prompts. The sensor network is responsible for collecting environmental information behind the vehicle, and the algorithm model analyzes and processes this information to determine whether there are safety hazards and issue early warning signals when necessary.

[0004] However, the existing technology still has some shortcomings in practical applications. For example, the coverage and accuracy of the sensor network are limited, which may cause some safety hazards to be missed; the complexity and computational complexity of the algorithm model are large, which may affect the real-time performance and accuracy of the system. Therefore, a new vehicle side rear safety warning technology is needed to solve these problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a side and rear visual safety warning method based on deep learning, and solves the problems raised in the above-mentioned background technology through the following scheme.

[0006] To achieve the above object, the present invention provides the following technical solution: a side and rear visual safety warning method based on deep learning, comprising: Step 1: Time zone division: used to determine the side and rear monitoring time of the target vehicle as the target time zone, divide the target time zone into various sub-time zones by equal time division, and mark them as 1, 2...n in sequence; Step 2: Lateral and rearward data collection: used to collect environmental perception data, target feature data, optical imaging data, and system performance data in each sub-time zone, and transmit the collected data to step 3; Step 3: Lateral and rearward data analysis: used to establish environmental perception data analysis methods, target feature data analysis methods, optical imaging data analysis methods, and system performance data analysis methods, used to analyze the data transmitted in step 2, and transmit the analysis results to step 4; Step 4: Comprehensive analysis: used to establish a comprehensive analysis model, conduct a comprehensive analysis on the data transmitted in step 3, and transmit the analysis results to step 5; Step 5: Safety warning: It is used to establish a threshold value of the comprehensive abnormal index of the side and rear, make an abnormal judgment on the target vehicle according to the threshold value of the comprehensive abnormal index of the side and rear, and issue a warning signal.

[0007] Preferably, the environmental perception data include the ambient light intensity attenuation coefficient, the atmospheric visibility index, the air particle concentration distribution and the ambient reflectivity coefficient, which are marked as El, Ea, Ep and Ec respectively; the target feature data include the target surface material reflection coefficient, the target contour edge gradient value, the target thermal radiation intensity and the target motion vector field density, which are marked as Sm, Se, St and Sv respectively; the optical imaging data include the lens distortion coefficient matrix, the spatial resolution attenuation rate, the imaging system modulation transfer function value and the spectral response curve integral value, which are marked as Ld, Ls, Lm and Lr respectively; the system performance data include the real-time processing delay time, the target detection confidence, the system anti-interference ability coefficient and the multi-target tracking accuracy, which are marked as Rp, Rd, Rc and Rm respectively.

[0008] Preferably, the ambient light intensity attenuation coefficient uses a photometer array to measure the light intensity at different distance points and is calculated through an exponential decay model. The atmospheric visibility index is obtained using a visibility meter through the scattered light measurement principle, or a laser rangefinder for multi-point measurement. The air particle concentration distribution is monitored in real time by a laser scattering particle sensor and multi-point sampling is averaged. The ambient reflectance coefficient is measured by a spectrophotometer to measure the ratio of incident light and reflected light intensities.

[0009] Preferably, the target surface material reflectance coefficient is collected in real time by a bidirectional reflectance distribution function measuring instrument, the target contour edge gradient value is captured by a high-speed camera, the edge gradient is calculated in real time using a Sobel operator, the target thermal radiation intensity is measured by an infrared thermal imager array to measure the target surface temperature distribution, and the target motion vector field density is calculated by a multi-camera array optical flow method to calculate the motion vector field.

[0010] Preferably, the lens distortion coefficient matrix calculates distortion parameters by using a calibration plate in conjunction with Zhang Zhengyou calibration method, the spatial resolution attenuation rate is measured by using a resolution test card to measure MTF value changes at different distances, the imaging system modulation transfer function value is measured by using a sine test card to measure different spatial frequency responses, and the integral value of the spectral response curve is measured by scanning the full spectral response curve with a spectrophotometer.

[0011] Preferably, the real-time processing delay time is obtained by recording the signal input and output time difference through a high-precision timestamp, the target detection confidence is obtained through the probability value output by the deep learning model, the system anti-interference ability coefficient is measured by a signal-to-noise ratio meter to measure the system's performance under different interference intensities, and the multi-target tracking accuracy is calculated by comparing with the ground truth value. CLEARMOT index.

[0012] Preferably, the environmental perception data analysis method is used to establish an environmental perception data analysis model for analyzing environmental perception data, which is specifically expressed as follows: , E i represents the environmental perception evaluation value of the i-th sub-time zone, El i Represents the ambient light intensity attenuation coefficient of the i-th sub-time zone, Ea i Represents the atmospheric visibility index of the ith sub-time zone, Ep i represents the distribution of air particle concentration in the ith sub-time region, Ec i Represents the ambient reflectance coefficient of the i-th sub-time region.

[0013] Preferably, the target feature data analysis method is used to establish a target feature data analysis model for analyzing the target feature data, which is specifically expressed as: , S i represents the target feature evaluation value of the i-th sub-time region, Sm i represents the target surface material reflection coefficient in the i-th sub-time zone, Se i St represents the edge gradient value of the target contour in the i-th sub-time region, i represents the target thermal radiation intensity in the ith sub-time zone, Sv i Represents the target motion vector field density of the i-th sub-time region.

[0014] Preferably, the optical imaging data analysis method is used to establish an optical imaging data analysis model for analyzing optical imaging data, which is specifically expressed as follows: , L i represents the optical imaging data evaluation value of the ith sub-time region, Ld i represents the lens distortion coefficient matrix of the i-th sub-time region, Ls i represents the spatial resolution decay rate of the ith sub-time region, Lm i represents the modulation transfer function value of the imaging system in the i-th sub-time region, Lr i Represents the integral value of the spectral response curve in the i-th sub-time region.

[0015] Preferably, the system performance data analysis method is used to establish a system performance data analysis model for analyzing system performance data, which is specifically expressed as follows: , Ri represents the system performance evaluation value of the ith sub-time zone, Rp i represents the real-time processing delay time of the i-th sub-time zone, Rd i represents the target detection confidence of the i-th sub-time region, Rc i Represents the system anti-interference capability coefficient of the i-th sub-time zone, Rm i Represents the multi-target tracking accuracy in the i-th sub-time region.

[0016] Preferably, the comprehensive analysis model is specifically expressed as: , σ represents the comprehensive abnormal index of the target vehicle’s side and rear, E i represents the environmental perception evaluation value of the i-th sub-time zone, S i represents the target feature evaluation value of the i-th sub-time region, L i represents the optical imaging data evaluation value of the ith sub-time region, R i represents the system performance evaluation value of the ith sub-time zone, and µ represents other influencing factors of the lateral and rearward comprehensive abnormality index.

[0017] Preferably, the lateral and rear comprehensive abnormality index threshold is specifically expressed as σ0. When σ0≥σ, it indicates that there is no visual abnormality in the lateral and rearward direction of the target vehicle, indicating monitoring of the target vehicle. When σ0<σ, it indicates that there is a visual abnormality in the lateral and rearward direction of the target vehicle, and a warning signal is issued to the management personnel terminal.

[0018] Technical effects and advantages of the present invention: The present invention determines the side and rear monitoring time of the target vehicle as the target time zone by time zone division, and divides the time into each sub-time zone, so as to realize the refined management of the monitoring time. This division is helpful for the continuity and accuracy of subsequent data collection and analysis, and provides a reliable time basis for safety warning; the environmental perception data, target feature data, optical imaging data and system performance data of each sub-time zone are collected by side and rear data, so as to fully reflect the actual situation of the side and rear of the target vehicle, and these data provide a rich information source for subsequent data analysis, which is helpful for discovering potential safety hazards; the environmental perception data analysis method, the target feature data analysis method, the optical imaging data analysis method and the system performance data analysis method are established by side and rear data analysis, so as to conduct in-depth analysis and processing on the collected data, and these methods can extract key information from the data, and provide strong support for comprehensive analysis and safety warning; the comprehensive analysis model is established by comprehensive analysis, and the data transmitted in step 3 are comprehensively analyzed, so as to obtain the comprehensive abnormal index of the side and rear of the target vehicle, which comprehensively considers factors such as environmental perception, target feature, optical imaging and system performance, and can fully reflect the safety status of the side and rear of the target vehicle. Through comprehensive analysis, potential safety risks can be discovered and warned in a timely manner; through safety warning, a threshold value of the lateral and rear comprehensive abnormality index is established, and the target vehicle is judged to be abnormal based on the threshold. When the lateral and rear comprehensive abnormality index exceeds the threshold, the system can automatically send a warning signal to the management terminal. This warning mechanism can timely discover and remind management personnel to pay attention to the lateral and rear safety issues of the target vehicle, which helps to take timely and effective measures to avoid potential safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] refer to Figure 1 The side and rear visual safety warning method based on deep learning shown in the figure comprises the following specific steps: Step 1: Time zone division: used to determine the side and rear monitoring time of the target vehicle as the target time zone, and divide the target time zone into various sub-time zones by equal time division, and mark them as 1, 2...n in sequence.

[0022] Step 2: Lateral and rearward data collection: used to collect environmental perception data, target feature data, optical imaging data, and system performance data for each sub-time zone, and transmit the collected data to step 3.

[0023] The environmental perception data include the ambient light intensity attenuation coefficient, the atmospheric visibility index, the air particle concentration distribution and the ambient reflectivity coefficient, which are marked as El, Ea, Ep and Ec respectively; the target feature data include the target surface material reflection coefficient, the target contour edge gradient value, the target thermal radiation intensity and the target motion vector field density, which are marked as Sm, Se, St and Sv respectively; the optical imaging data include the lens distortion coefficient matrix, the spatial resolution attenuation rate, the imaging system modulation transfer function value and the spectral response curve integral value, which are marked as Ld, Ls, Lm and Lr respectively; the system performance data include the real-time processing delay time, the target detection confidence, the system anti-interference ability coefficient and the multi-target tracking accuracy, which are marked as Rp, Rd, Rc and Rm respectively.

[0024] The ambient light intensity attenuation coefficient uses a photometer array to measure the light intensity at different distance points and is calculated through an exponential decay model. The atmospheric visibility index is obtained using a visibility meter through the scattered light measurement principle, or a laser rangefinder for multi-point measurement. The air particle concentration distribution is monitored in real time by a laser scattering particle sensor, and multi-point sampling is averaged. The ambient reflectance coefficient is measured by a spectrophotometer to measure the ratio of incident light to reflected light intensity.

[0025] The target surface material reflectance coefficient is collected in real time by a bidirectional reflectance distribution function measuring instrument, the target contour edge gradient value is captured by a high-speed camera, and the edge gradient is calculated in real time using the Sobel operator, the target thermal radiation intensity is measured by an infrared thermal imager array to measure the target surface temperature distribution, and the target motion vector field density is calculated by a multi-camera array optical flow method.

[0026] The lens distortion coefficient matrix is ​​calculated by using a calibration plate in conjunction with Zhang Zhengyou's calibration method to calculate distortion parameters, the spatial resolution attenuation rate is measured by using a resolution test card to measure the MTF value change at different distances, the imaging system modulation transfer function value is measured by using a sine test card to measure different spatial frequency responses, and the spectral response curve integral value is measured by scanning the full spectral response curve with a spectrophotometer.

[0027] The real-time processing delay time is recorded by a high-precision timestamp to record the signal input and output time difference, the target detection confidence is the probability value output by the deep learning model, the system anti-interference ability coefficient is measured by a signal-to-noise ratio meter to measure the performance of the system under different interference intensities, and the multi-target tracking accuracy is calculated by comparing with the ground truth value. CLEAR MOT index.

[0028] Step 3: Lateral and rearward data analysis: used to establish environmental perception data analysis methods, target feature data analysis methods, optical imaging data analysis methods, and system performance data analysis methods, used to analyze the data transmitted in step 2, and transmit the analysis results to step 4.

[0029] The environmental perception data analysis method is used to establish an environmental perception data analysis model for analyzing environmental perception data, which is specifically expressed as follows: , E i represents the environmental perception evaluation value of the i-th sub-time zone, El i Represents the ambient light intensity attenuation coefficient of the i-th sub-time zone, Ea i Represents the atmospheric visibility index of the ith sub-time zone, Ep i represents the distribution of air particle concentration in the ith sub-time region, Ec i Represents the ambient reflectance coefficient of the i-th sub-time region.

[0030] The target feature data analysis method is used to establish a target feature data analysis model for analyzing target feature data, which is specifically expressed as follows: , S i represents the target feature evaluation value of the i-th sub-time region, Sm i represents the target surface material reflection coefficient in the i-th sub-time zone, Se i St represents the edge gradient value of the target contour in the i-th sub-time region, i represents the target thermal radiation intensity in the ith sub-time zone, Sv i Represents the target motion vector field density of the i-th sub-time region.

[0031] The optical imaging data analysis method is used to establish an optical imaging data analysis model for analyzing optical imaging data, which is specifically expressed as follows: , L i represents the optical imaging data evaluation value of the ith sub-time region, Ld i represents the lens distortion coefficient matrix of the i-th sub-time region, Ls i represents the spatial resolution decay rate of the ith sub-time region, Lm i represents the modulation transfer function value of the imaging system in the i-th sub-time region, Lr i Represents the integral value of the spectral response curve in the i-th sub-time region.

[0032] The system performance data analysis method is used to establish a system performance data analysis model for analyzing system performance data, which is specifically expressed as follows: , R i represents the system performance evaluation value of the ith sub-time zone, Rp irepresents the real-time processing delay time of the i-th sub-time zone, Rd i represents the target detection confidence of the i-th sub-time region, Rc i Represents the system anti-interference capability coefficient of the i-th sub-time zone, Rm i Represents the multi-target tracking accuracy in the i-th sub-time region.

[0033] Step 4: Comprehensive analysis: used to establish a comprehensive analysis model, conduct a comprehensive analysis on the data transmitted in step 3, and transmit the analysis results to step 5.

[0034] The comprehensive analysis model is specifically expressed as: , σ represents the comprehensive abnormal index of the target vehicle’s side and rear, E i represents the environmental perception evaluation value of the i-th sub-time zone, S i represents the target feature evaluation value of the i-th sub-time region, L i represents the optical imaging data evaluation value of the ith sub-time region, R i represents the system performance evaluation value of the ith sub-time zone, and µ represents other influencing factors of the lateral and rearward comprehensive abnormality index.

[0035] Step 5: Safety warning: It is used to establish a threshold value of the comprehensive abnormal index of the side and rear, make an abnormal judgment on the target vehicle according to the threshold value of the comprehensive abnormal index of the side and rear, and issue a warning signal.

[0036] The lateral and rearward comprehensive abnormality index threshold is specifically expressed as σ0. When σ0≥σ, it indicates that there is no abnormality in the lateral and rearward visibility of the target vehicle, indicating monitoring of the target vehicle. When σ0<σ, it indicates that there is an abnormality in the lateral and rearward visibility of the target vehicle, and a warning signal is sent to the management terminal.

[0037] The present invention determines the side and rear monitoring time of the target vehicle as the target time zone by time zone division, and divides the time into each sub-time zone, so as to realize the refined management of the monitoring time. This division is helpful for the continuity and accuracy of subsequent data collection and analysis, and provides a reliable time basis for safety warning; the environmental perception data, target feature data, optical imaging data and system performance data of each sub-time zone are collected by side and rear data, so as to fully reflect the actual situation of the side and rear of the target vehicle, and these data provide a rich information source for subsequent data analysis, which is helpful for discovering potential safety hazards; the environmental perception data analysis method, the target feature data analysis method, the optical imaging data analysis method and the system performance data analysis method are established by side and rear data analysis, so as to conduct in-depth analysis and processing on the collected data, and these methods can extract key information from the data, and provide strong support for comprehensive analysis and safety warning; the comprehensive analysis model is established by comprehensive analysis, and the data transmitted in step 3 are comprehensively analyzed, so as to obtain the comprehensive abnormal index of the side and rear of the target vehicle, which comprehensively considers factors such as environmental perception, target feature, optical imaging and system performance, and can fully reflect the safety status of the side and rear of the target vehicle. Through comprehensive analysis, potential safety risks can be discovered and warned in a timely manner; through safety warning, a threshold value of the lateral and rear comprehensive abnormality index is established, and the target vehicle is judged to be abnormal based on the threshold. When the lateral and rear comprehensive abnormality index exceeds the threshold, the system can automatically send a warning signal to the management terminal. This warning mechanism can timely discover and remind management personnel to pay attention to the lateral and rear safety issues of the target vehicle, which helps to take timely and effective measures to avoid potential safety accidents.

[0038] Secondly: In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other; Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A side and rear visual safety warning method based on deep learning, characterized in that: include: Step 1: Time zone division: used to determine the side and rear monitoring time of the target vehicle as the target time zone, divide the target time zone into various sub-time zones by equal time division, and mark them as 1, 2...n in sequence; Step 2: Lateral and rearward data collection: used to collect environmental perception data, target feature data, optical imaging data, and system performance data in each sub-time zone, and transmit the collected data to step 3; Step 3: Lateral and rearward data analysis: used to establish environmental perception data analysis methods, target feature data analysis methods, optical imaging data analysis methods, and system performance data analysis methods, used to analyze the data transmitted in step 2, and transmit the analysis results to step 4; Step 4: Comprehensive analysis: used to establish a comprehensive analysis model, conduct a comprehensive analysis on the data transmitted in step 3, and transmit the analysis results to step 5; Step 5: Safety warning: It is used to establish a threshold value of the comprehensive abnormal index of the side and rear, make an abnormal judgment on the target vehicle according to the threshold value of the comprehensive abnormal index of the side and rear, and issue a warning signal.

2. The side and rear visual safety warning method based on deep learning according to claim 1 is characterized by: The environmental perception data include the ambient light intensity attenuation coefficient, the atmospheric visibility index, the air particle concentration distribution and the ambient reflectivity coefficient, which are marked as El, Ea, Ep and Ec respectively; the target feature data include the target surface material reflection coefficient, the target contour edge gradient value, the target thermal radiation intensity and the target motion vector field density, which are marked as Sm, Se, St and Sv respectively; the optical imaging data include the lens distortion coefficient matrix, the spatial resolution attenuation rate, the imaging system modulation transfer function value and the spectral response curve integral value, which are marked as Ld, Ls, Lm and Lr respectively; the system performance data include the real-time processing delay time, the target detection confidence, the system anti-interference ability coefficient and the multi-target tracking accuracy, which are marked as Rp, Rd, Rc and Rm respectively.

3. The side and rear visual safety warning method based on deep learning according to claim 2 is characterized by: The ambient light intensity attenuation coefficient uses a photometer array to measure the light intensity at different distance points and is calculated through an exponential decay model. The atmospheric visibility index is obtained using a visibility meter through the scattered light measurement principle, or a laser rangefinder for multi-point measurement. The air particle concentration distribution is monitored in real time by a laser scattering particle sensor, and multi-point sampling is averaged. The ambient reflectance coefficient spectrophotometer measures the ratio of incident light to reflected light intensity. The target surface material reflectance coefficient is collected in real time by a bidirectional reflectance distribution function measuring instrument, the target contour edge gradient value is captured by a high-speed camera, and the edge gradient is calculated in real time using the Sobel operator, the target thermal radiation intensity is measured by an infrared thermal imager array to measure the target surface temperature distribution, and the target motion vector field density is calculated by a multi-camera array optical flow method. The lens distortion coefficient matrix is ​​calculated by using a calibration plate in conjunction with the Zhang Zhengyou calibration method to calculate the distortion parameters, the spatial resolution attenuation rate is measured by using a resolution test card to measure the MTF value change at different distances, the imaging system modulation transfer function value is measured by using a sine test card to measure different spatial frequency responses, and the spectral response curve integral value is measured by scanning the full spectral response curve with a spectrophotometer; The real-time processing delay time is recorded by a high-precision timestamp to record the signal input and output time difference, the target detection confidence is the probability value output by the deep learning model, the system anti-interference ability coefficient is measured by a signal-to-noise ratio meter to measure the performance of the system under different interference intensities, and the multi-target tracking accuracy is calculated by comparing with the ground truth value. CLEAR MOT index.

4. The side and rear visual safety warning method based on deep learning according to claim 1 is characterized by: The environmental perception data analysis method is used to establish an environmental perception data analysis model for analyzing environmental perception data, which is specifically expressed as follows: , E i represents the environmental perception evaluation value of the i-th sub-time zone, El i Represents the ambient light intensity attenuation coefficient of the i-th sub-time zone, Ea i Represents the atmospheric visibility index of the ith sub-time zone, Ep i represents the distribution of air particle concentration in the ith sub-time region, Ec i Represents the ambient reflectance coefficient of the i-th sub-time region.

5. The side and rear visual safety warning method based on deep learning according to claim 1 is characterized by: The target feature data analysis method is used to establish a target feature data analysis model for analyzing target feature data, which is specifically expressed as follows: , S i represents the target feature evaluation value of the i-th sub-time region, Sm i represents the target surface material reflection coefficient in the i-th sub-time zone, Se i St represents the edge gradient value of the target contour in the i-th sub-time region, i represents the target thermal radiation intensity in the ith sub-time zone, Sv i Represents the target motion vector field density of the i-th sub-time region.

6. The side and rear visual safety warning method based on deep learning according to claim 1 is characterized by: The optical imaging data analysis method is used to establish an optical imaging data analysis model for analyzing optical imaging data, which is specifically expressed as follows: , L i represents the optical imaging data evaluation value of the ith sub-time region, Ld i represents the lens distortion coefficient matrix of the i-th sub-time region, Ls i represents the spatial resolution decay rate of the ith sub-time region, Lm i represents the modulation transfer function value of the imaging system in the i-th sub-time region, Lr i Represents the integral value of the spectral response curve in the i-th sub-time region.

7. The side and rear visual safety warning method based on deep learning according to claim 1 is characterized by: The system performance data analysis method is used to establish a system performance data analysis model for analyzing system performance data, which is specifically expressed as follows: , R i represents the system performance evaluation value of the ith sub-time zone, Rp i represents the real-time processing delay time of the i-th sub-time zone, Rd i represents the target detection confidence of the i-th sub-time region, Rc i Represents the system anti-interference capability coefficient of the i-th sub-time zone, Rm i Represents the multi-target tracking accuracy in the i-th sub-time region.

8. The side and rear visual safety warning method based on deep learning according to claim 1 is characterized by: The comprehensive analysis model is specifically expressed as: , σ represents the comprehensive abnormal index of the target vehicle’s side and rear, E i represents the environmental perception evaluation value of the i-th sub-time zone, S i represents the target feature evaluation value of the i-th sub-time region, L i represents the optical imaging data evaluation value of the ith sub-time region, R i represents the system performance evaluation value of the ith sub-time zone, and µ represents other influencing factors of the lateral and rearward comprehensive abnormality index.

9. The side and rear visual safety warning method based on deep learning according to claim 1 is characterized by: The lateral and rearward comprehensive abnormality index threshold is specifically expressed as σ0. When σ0≥σ, it indicates that there is no abnormality in the lateral and rearward visibility of the target vehicle, indicating monitoring of the target vehicle. When σ0<σ, it indicates that there is an abnormality in the lateral and rearward visibility of the target vehicle, and a warning signal is sent to the management terminal.

Images