Research On Sensor Sharing Strategy For Perception Blind Spot Supplement In Intelligent Driving

Intelligent driving is a kind of technology which can help people to drive and it can completely replace human driving in special circumstances as well as it is one of the important ways to improve road traffic safety.Specifically,when a variety of data fusion,intelligent driving can realize the vehicle’s accurate perception of the surrounding environment and it can obtain its own exact environment and realize the autonomous driving of the vehicle.However,vehicle sensors have their own limitations of perception.For instance,when the sensing spot is blocked by other objects,resulting in the perception blind spot.It would be a huge challenge to achieve intelligent driving with only a single on-board sensory device.Therefore,how to select the appropriate vehicle nodes that can be shared with vehicle sensors is an important research direction.This paper mainly focuses on the blind spot of sensors(such as camera,lidar and millimeter wave radar)with a certain depth and breadth of sensing range,and studies the sensor sharing strategy for the supplement of blind spot of intelligent driving perception.Firstly,in order to analyze the sensor blind spot supplement range,a sensor blind spot model in dynamic traffic flow is proposed.Secondly,a blind spot supplement strategy based on the selection of sensor shared nodes is proposed by considering the multi-dimensional indicators of candidate nodes.The specific research is as follows:(1)Aiming at the problem that the blind spot cannot be sensed by sensors due to the occlusion between vehicles,but the current sensor sharing field lacks the mathematical model for analyzing the blind spot sensed by sensors.To solve these problems,this paper uses the vehicle distribution model in dynamic traffic flow to analyze the relationship between vehicle spacing and vehicle density,and uses the location information of vehicles in dynamic traffic flow to select the most suitable distribution form and estimate distribution parameters.Moreover,based on the kinematics model,a safe distance model for intelligent driving at a constant speed was proposed,and the perception range required by the vehicle to ensure safety was defined.Finally,a sensor perception blind spot model in dynamic traffic flow is proposed.This model combines the vehicle distribution model and the safety distance model to define the sensor perception blind spot and the blind spot supplement,and theoretically analyzes the mapping relationship between the perception blind spot and the vehicle density.The experimental results show that the mathematical relationship between vehicle spacing and blind spot in dynamic traffic flow is well represented.(2)Aiming at the current node selection strategy,which ignores the problem of intelligent driving cars to make up for the blind spot,a blind spot supplement spot quantization algorithm is designed.The algorithm using the safety range of the current vehicle and the position information of the occlusion to determine the sensing blind spot range of the current vehicle sensor.Subsequently,based on the location of surrounding vehicles and their perception range data,geometric methods are used to quantify their supplementary spot for the blind spot of the current vehicle.Furthermore,for the problem that the current node selection strategy does not consider the impact of sensor data quality and communication system performance on the sharing quality,a sensor sharing node selection strategy based on entropy weight method is proposed based on the above algorithm.The strategy comprehensively considers together with the supplement the spot of the blind spot of the current vehicle by the surrounding vehicles,the data reliability index of the sensor and the performance index of communication that must be considered in intelligent driving.Then use the entropy weight method to compare and score the vehicles.Based on the score,the sensor sharing node selection that comprehensively considers multi-dimensional indicators is realized.The experimental results show that this strategy achieves a higher blind spot supplement ratio and can effectively compare multi-dimensional indicators.