Research On Data Fusion Algorothm Under Multi-Sensor Redundancy For Driverless Cars

With the development of automobile electrification and intelligence,autonomous driving systems are ushering in new development opportunities in the2020 s,but in the current technological iteration process,there have been many unexpected accidents,many of which are automatic driving perception systems.Because of the serious consequences caused by the failure of the system,the redundant fusion of the sensing system is imminent.This article proposes a new solution to the perception of autonomous vehicles by multi-sensor redundancy.The specific research content is as follows:Firstly,the target detection algorithm of Lidar,Camera,and Radar was studied.Among them,the Lidar selected CNN＿SEG algorithm which is used in Apollo,and the camera selected the YOLOV4 algorithm to detect the target.There was no Radar data in the data set,and the object detection in the world should met the Gaussian distribution,so the maximum and minimum values of x/y of the Lidar object were processed by Box-Muller algorithm to obtain the target as the Radar target.The problem of the target detection part was solved.For a multi-sensor system,the most important thing to do was to align the coordinate system of each sensor uniformly.The data set used by this system was the Kitti data set.Under this data set,the coordinates of the Lidar contacted to the color camera.The transformation required a rotation matrix and a translation matrix.With these two matrices,the coordinate system of the two sensors can be transformed.The camera coordinate system was selected as the main coordinate system,and the Lidar object and Radar object were transformed to the camera coordinates.Under the system,the coordinate transformation problem was completed.In order to evaluate the detection status of each sensor,the multi-target tracking of the sensor was very necessary.In the sensor tracking part,the SORT algorithm was adopted.The algorithm included Hungarian matching data association and Kalman filter trajectory tracking.The multi-target of the sensor was calculated using the SORT algorithm,and the obtained multi-target variance was used for the subsequent multi-sensor redundant weight distribution.The use of dynamic weight allocation mechanism and sensor elimination mechanism can greatly reduced the impact of sensor tracking faults on multi-sensor sensing systems.This paper developed a Dynamic Weight Distribution algorithm—DWD algorithm,which can minimized traffic accident caused by a sensor sensing system failure.Through dynamic weight distribution,during the operation of the multi-sensor fusion system,the weight of sensors with poor tracking performance was actively reduced.Through the threshold setting,the object of the sensor that had a particularly poor tracking effect was eliminated.The settings of the above two mechanisms ensured the stability of the fusion data and ensured the smooth operation of the fusion system.By combining the redundant weight of multi-sensor with the sensor information fusion algorithm,the multi-sensor target was adapt redundant,and the redundant weight of each sensor was adjusted in real time during the operation of the system,thereby affecting the fused target accuracy,if there was an obvious problem in the target tracking of one or more sensors,the sensor will be withdraw from the fusion,that was,it will not participate in the sensor fusion,and it will not believe in the detected target,so as to avoid some malfunctions due to the sensor itself.The perception errors caused by the perception errors can then avoid some traffic accidents caused by perception errors and provide a more solid guarantee for the safety of people’s lives and property.