Date of Award

5-2018

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

Committee Chair

Voicu Popescu

Committee Co-Chair

Jiang Yu Zheng

Committee Member 1

Xavier Tricoche

Committee Member 2

Mihran Tuceryan

Committee Member 3

Lauren Christopher

Abstract

Nowadays, many vehicles are equipped with a vehicle borne camera system for monitoring drivers’ behavior, accident investigation, road environment assessment, and vehicle safety design. This produces a huge amount of video data recorded daily. Analyzing and interpreting these data in an efficient way has become a non-trivial task. Therefore, in this work, for efficient analysis of these data, videos are mapped into a temporal profile image of reduced dimension for acquiring motion information. The video profile is compact and continues motion representation of a video. This work examines those motion profile trajectories. A new motion based collision avoidance and pedestrian detection systems are proposed and developed. After extracting motion profiles, various driving collision and pedestrian walking scenarios are investigated in depth. New algorithms have been designed to compute flow divergence for collision warning, and detect walking pedestrians by their leg chains in the motion profiles. In other words, this work proposes a uniform framework for extraction of motion profiles from a video, instantaneous Time-to-Collision (TTC) computation, and leg-chain detections for pedestrian safety. The convolutional filtering technique has been used to acquire motion. The results can be interpreted directly in the motion profiles without requiring watching the video sequences.

Download

Share

COinS