Date of Award

8-2018

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering (MSEE)

Department

Electrical Engineering

First Advisor

Song, JainJain

Second Advisor

Yoder, Mark

Third Advisor

McInerney, Michael

Abstract

The objective of this thesis is to improve the accuracy of predicting motion trajectory, i.e., speed and direction, of a pedestrian in front of an Ego Vehicle which has a Mobileye camera with an advanced driver assistance system (ADAS). The Ego Vehicle captures and records videos of pedestrians in front of it, and these videos are analyzed to predict a pedestrian trajectory from instantaneous, random actions of a pedestrian. Instant actions include, but are not limited to, walking at a constant speed, sudden accelerations/decelerations, sudden dodging from the Ego Vehicle, sudden advancements to the Ego Vehicle, sudden withdrawals or sudden stops at the road edge, etc. Pedestrian positions and motion data from the videos can be used to estimate pedestrian state parameters and predict pedestrian movement. The pedestrian videos contain noises due to the nonlinear trajectory of a pedestrian and the Ego Vehicle. An extended Kalman filter (EKF) and pedestrian behavior classification are applied to these pedestrian videos to obtain a more accurate pedestrian trajectory. The EKF is used to suppress noises from the videos and aids in predicting the next state of pedestrian movement. The EKF can reduce noises in a nonlinear system. The EKF is an efficient and effective tool in creating more stable and smoother pedestrian positions from the Ego Vehicle videos, as we have demonstrated from analyzing pedestrian trajectories from real-world videos. These new position data inputs are used to calculate the new velocity of a pedestrian. This new velocity is averaged over 30 consecutive video frames to obtain a more accurate and stable velocity. After the new position and velocity are calculated, pedestrian behavior classification is applied to the data to calculate and group pedestrian behaviors into instant actions. The behavior classification is based on the estimation of the heading angle and acceleration of a pedestrian. The combination of the extended Kalman filter and behavior classification forms a more accurate pedestrian trajectory prediction system. This approach is verified with 12 hours of ADAS camera Mobileye videos from an experimental car test site within a simulated urban area. Ten cases of pedestrian motion behaviors are analyzed. By calculating the Time to Collision (TTC) and comparing this result with the TTC directly from the ADAS camera, we have shown that our new TTC prediction is more stable and less noisy when contrasted with the older TTC predictions from an ADAS camera system.

Download

Share

COinS