Analysis and Multi - target Tracking based on Dynamic Programming

WEI, CHUNPENG. Driver Assistant System: Robust Segmentation based on Topological Persistent Analysis and Multi-target Tracking based on Dynamic Programming. (Under the direction of Dr. Edgar Lobaton.) This thesis work mainly focus on two topics. First, a new methodology for robust segmentation of obstacles from stereo disparity maps in an on-road environment is presented. The thesis work first construct a probability of the occupancy grid map using the UV-disparity methodology. Traditionally, a simple threshold has been applied to segment the resulting region; however, this outcome is sensitive to the parameter value. Instead of thresholding, the thesis work perform a topological persistence analysis on the constructed occupancy map. The topological frame-work hierarchically encodes all possible segmentation results as a function of the threshold, thus can identify the regions that are most persistent. This leads to a more robust segmentation. Second, a robust multitarget tracking technique for obstacles on the road using depth imaging information is proposed. The novelty of this proposed approach lies in the use of the robust obstacle segmentation method based on topological persistence, as input to a max flow based tracking algorithm. To reduce time as well as computational complexity, the max flow problem is solved using a dynamic programming algorithm. This thesis work incorporate Kalman filter and neural network in this method to attain robust tracking results. The proposed tracking algorithm has been tested on several stereo datasets and the results show that there is an improvement on robustness when comparing performance with and without the topological persistent detection. © Copyright 2015 by Chunpeng Wei