Application Of Radar And Camera Data Fusion In Intelligent Driving Assistance

The rapid development and broad application prospects of intelligent driving technology put forward higher requirements for the performance of intelligent assistant driving system.Considering the shortcomings of single sensor environment sensing schemes which are widely used in mass production at present,the problems at this stage can be obviously solved by redundancy and complementarity of multi-sensor information.For this reason,this paper proposes a data fusion algorithm based on target level of radar and camera,and designs a fusion framework based on decision level data fusion idea,including fusion data preprocessing,target level data fusion,lane line data fusion and the most dangerous target screening strategy.In radar raw data,the effective target selection algorithm and effective life cycle test method designed in this paper can eliminate a large number of noise and interference signals.The inconsistency between the observed data of the camera and the detected target can be solved by continuous tracking of the original target of the camera and correlation of the target data.Kalman filtering based on constant acceleration motion model can significantly improve the quality of the original target signal.In target-level data fusion,improving the traditional European distance can effectively judge whether the target information observed independently by radar and camera is the observation of the same actual detected target.Kalman weighted fusion algorithm can synthesize the advantages of radar target and camera target observation data and obtain fusion target information with high accuracy and small error.Continuous multi-frame data tracking of fused targets can effectively solve the problem of single sensor target loss and reduce the false detection rate of the target in front of the environment sensing system.Based on the historical track information of the stable vehicle ahead and the cubic curve fitting of the lane line of the road ahead by the least square method,the lane information with high reliability can be obtained in lane data fusion.The lane line fusion strategy is designed for the three independent sources of the road information,which improves the adaptability of the environmental perception system.In the most dangerous target screening strategy,this paper divides the reasonable hazard judgment area of the front vehicle.By judging the lateral motion parameters of the front vehicle,the most dangerous target information can be obtained effectively and timely.A joint simulation platform of PreScan and Matlab/Simulink is built in this paper.In PreScan's simulation environment,the decision-level fusion algorithm is simulated and analyzed by simulating various complex scenes on real roads and the influence of weather on millimeter-wave radar sensors.The decision-level fusion algorithm designed in this paper is preliminarily verified and optimized,including simulation verification of car-following,downhill and dangerous cut-in scenarios.Because the target-level data observed by sensors in the simulation environment can not well reflect the actual situation,this paper builds a real vehicle verification environment with the help of available experimental conditions.Through repeated tests of various actual social road conditions,the decision-level fusion algorithm designed in this paper is debugged,tested and optimized comprehensively and systematically.Through a large number of simulation tests and real vehicle tests,the decision-level fusion algorithm based on target-level data of millimeter wave radar and camera designed in this paper can solve the problems of single sensor target detection.The reliability,stability and adaptability of environmental sensing system are improved by data redundancy and complementary advantages of sensors.At present,the hot intelligent driving technology is developing.Under the background,it has certain application value and research significance.