Risk Assessment Of The Front Vehicles For Intelligent Connected Vehicles

Intelligent network vehicle is an important development direction of the cuttingedge science and technology,which has a significant impact on intelligent transportation,modern logistics and other aspects.The environmental awareness system is the foundation and guarantee of the safety and intelligence of the intelligent networked vehicle.The fusion of radar and camera can detect the vehicle in front of it,and get the information of position,distance,speed and so on.Through these information,the possibility and danger degree of potential collision between the vehicle and the vehicle are analyzed,and different risk levels are divided.This risk level can be used as the input of intelligent decision-making system,so as to make more safe and effective driving decisions.Therefore,the main contents of this paper are as follows:(1)Multi radar data fusionAiming at the complementarity and redundancy of multi radar data,this paper proposes a method of multi radar data fusion.In view of the complementarity of multi radar data,a diamond grid map is proposed.Compared with other grid maps,the diamond grid map is closer to the area detected by multi radar in this paper.Aiming at the redundancy of multi radar data,a multi radar data fusion method based on dynamic parameter Bayes is proposed.The method of this paper can effectively reduce the radar miss detection rate and false alarm rate.(2)Vehicle detection based on fusion of radar and cameraThrough multi radar data fusion,the position information of the vehicle in front is obtained,and the position information of the vehicle in front is projected to the image collected by the camera through the joint calibration relationship between the radar and the camera to get the region of interest.In this area,the vehicle detection based on Alex net is carried out.The multi-sensor fusion + Alex used in this paper Net method has 94.5% accuracy rate,95.1% recall rate and 90 ms detection time,which basically meets the accuracy and real-time requirements of the system.(3)Risk level assessment of vehicles aheadFirstly,this paper collects the natural driving data from different drivers in the urban road environment of China for more than 50 hours,and the driving distance is nearly 1800 km.The data includes the driving status of the vehicle,the vehicle information in front of the potential collision and the brake line adopted by the driver.On this basis,the risk level is quantified by clustering the driver braking data after feature extraction.At the beginning of clustering,the number of clustering categories needs to be determined.In this paper,cfsfdp algorithm is used to determine the number of clustering categories.In addition,the data of different risk levels are unbalanced.In this paper,fuzzy c-harmonic mean with volume parameters is used to cluster.Next,this quantified risk level is used as the risk level label set of the vehicle ahead,and the risk level of the target vehicle ahead is evaluated by the method of random forest.The results show that the risk level of the proposed algorithm is divided into five levels,and the accuracy of risk level assessment for the target vehicle ahead is 94.3%.