Wrong Detection and Prediction Issue

[tiandaji]

As shown in the screenshot, when using object_tracker --car, there are wrong detection and prediction for the vehicles (in fact, the persons around the lidar carrier are recognized as car).

Therefore, I would like to know if it is possible to train the system with KITTI data, and if so, how to perform the process in order to improve the detection and prediction. Also will it be able to run both pedestrian and vehicle detection at the same time?