Target vehicle selection based on multi features fusion method

Abstract

A target selection method based on multi features fusion is proposed to improve the accuracy of target vehicle selection. The parameters consisting of the longitudinal distance, lateral distance, relative speed between objects and the host vehicle, the in-lane probability of objects are regarded as the features of individual vehicles. Firstly, some pre-processes of features data are carried out including Distance Compensation Factor (DCF) correcting and Kalman filtering, which are used to correct the in-lane probability data provided by lidar, track and predict the relative distance and speed of objects to lower the missing rate of vehicle detection respectively. Furthermore a two-layer BP neural network is designed to train the sample data and obtain the importance weight of feature variables; the training output is finally utilized as the index for target recognition. The selection method utilizes the valid information collected by sensors through the fusion of multi vehicle features. Experiments show that the vehicle detection results can be improved and the target selection and tracking accurately can be fulfilled through the proposed method. Even under cut-in conditions, the target can also be switched to the cut-in vehicle in time.