Research On Object Tracking And Location Estimation Based On Multi-Sensors Fusion

As more sensors are mounted on self-driving cars,the identifiable driving scenes can be more and more complicated.However,failure of the sensors will increase the safety risk of the automatic driving system.In order to effectively reduce the risk caused by this situation,different levels of multi-sensor fusion methods have been proposed.Effective multi-sensor fusion provides multi-dimensional environment data,which provides a new solution for target perception task.Object tracking and target location are focus on estimating the state and location about target in the environment.People have long been committed to researching and improving vision-based or radar-based object tracking algorithms.In contrast,object tracking algorithms based on fused data are still relatively few.The main directions of multi-sensor fusion include data alignment,spatial position calibration,and heterogeneous data fusion.The Kalman filter can be used to fuse heterogeneous sensor data related to motion state.The object tracking algorithm makes one-to-one correlation for the vary states of the target at different time points,and then according the Kalman filter,the latest state of the trajectory can be estimated.The continuity and accuracy of the target position trajectory are used as correlation evaluation criteria.The result of object tracking will always be subject to influences by the physical characteristics and accuracy of the sensor.The research results of this paper rely on the robot research group of Jilin University to carry out research on object tracking and location estimation technology based on multi-sensor fusion.From the perspective of project requirements,combined with the existing research results and the experiment platform of the research group,this paper proposes an object tracking and localization estimation method based on multi-sensor fusion technology.This method achieves the matching of radar and camera targets,improves the reliability of the original radar tracking results,completes the target location estimate and generates a local target map.The main contributions of this article include the following:1.This paper proposes a correlation matching algorithm for vehicle and pedestrian targets.Based on multi-sensor fusion technology,the camera target and radar target are matched accurately.The location information and feature information of the sensor target are combined to complete the heterogeneous target data.The exact match makes the sensor fusion results easier to be combined with the object tracking algorithm.2.In this paper,we try to track the goal by multi-sensor information fusion.Based on the existing object tracking algorithm,the Kalman filter model is redesigned based on the motion characteristics of vehicles and pedestrians respectively.The GNSS/IMU navigation information is added as motion information compensation,so that the tracking results obtained by calculation are more accurate.3.A confidence calculation method is proposed to determine the credibility between our tracking result and the original radar result to better the fusion result.According to the GNSS/IMU navigation information,we estimated the tracking result and drew the local target map.A few groups of real-time position information for the targets have been introduced as the contrast true value in order to get a more objective evaluation for the effect of the algorithm in this paper.Through the comparison experiments of various scenarios,it is verified that the target matching method proposed in this paper provides classification information and size data for the tracking results,which improves the accuracy of target matching.Combined with the improvement on the object tracking algorithm,the final experimental results,compared with the original tracking results of the radar,are improved effectively,which brings down the wrong identifications.The experimental results import a local target map as a visualization tool to better demonstrate the spatial relationship between the tracked target and the vehicle itself.