Object tracking and state estimation in outdoor scenes is crucial for autonomous vehicle research. For well fulfilling this mission, there are three key challenges should be overcome-obstacles detection and segmentation, ego-motion estimation and object tracking. In our method, 2D Gaussian process algorithm based on local samples is presented here for obstacle area detection. Then, obstacle area is further classified as different obstacles based on grid map. Finally, the tracking step is based on an assumption that all obstacles including static ones are moving objects relative to the sensor. The benefit of this assumption is that ego-motion of autonomous vehicle does not need to be estimated before tracking but after tracking. In this step, the Kalman Filter and iterative closed point (ICP) registration method are adopted.