Research On Target Tracking Method Based On Multi-source Information Fusion

Intelligent Driving Assistance System has become one of the key technologies to improve the driving safety of vehicles,among which the sensor technology for environmental awareness has attracted more and more attention.Due to the limitation of production technology and its own conditions,the single sensor sometimes misses and misdetects the target in the actual application process,which will seriously interfere with the effective judgment of the target by the system.Therefore,the information fusion between the advantages of the combined sensors technology has become a hot topic for scholars at home and abroad.In this paper,the combination of millimeter wave radar and camera is used as an environmental information detection tool.A distributed fusion structure is established.Through collecting,detecting and identifying the target data,simultaneous interpreting and merging the trajectories from different sensors to obtain the target track,providing real-time and accurate driving environment information for the driver's active collision avoidance decision-making.The research contents are as follows:(1)Screen millimeter-wave radar data to determine effective targets.This paper first preprocesses the radar data,mainly screening and eliminating all zero value signals,static targets and invalid targets.Then,combined with the judgment method of the effective target of the contract lane,the non-dangerous target of the adjacent lane is filtered out to obtain the primary target.Finally,the validity of the existence of the target is further verified by the life cycle test method,and a Kalman filter model is established to filter the data information to reduce the noise interference.(2)Research on forward vehicle detection method based on machine vision.This paper uses the Haar-like method to extract vehicle features,while using the integration method to speed up the calculation of feature values.Based on these eigenvalues,several weak classifiers are trained by combining the algorithm iteratively,and the strong classifiers are constructed by combining them in series.Considering the continuity of the acquired vehicle images in time and space,this paper builds a vehicle tracking module based on the unscented Kalman filter algorithm to enhance the system's tracking of the target vehicle trajectory.(3)Establish a multi-sensor fusion model.By using the spatial position conversion relationship among pixel coordinate system,image coordinate system,camera coordinate system,world coordinate system and millimeter wave radar coordinate system to realize the spatial fusion of radar and camera.Aiming at the problem of different sampling frequency between millimeter wave radar and camera,this paper adopts multithreading method to realize time synchronization between sensors.Based on the principle of unscented information fusion and Extended information fusion,this paper proposes an improved unscented information fusion algorithm,which can eliminate the measurement information in the update equation of the fusion center,enhance the estimation accuracy of the target,and solve the problem that the target information among the sensors can not be effectively combined.(4)Simulation and real vehicle verification of the fusion system.In this paper,a six-degree-of-freedom QJ-4B1 dynamic simulation driver is used to simulate the fusion system.A virtual sensor,road conditions,and background environment are established through the human-computer interaction interface to maximize the reduction of real vehicle tests.In the real vehicle test,the installation of the sensor and the construction of the test platform were completed,and the real vehicle test site and personnel were arranged accordingly.