Human Detection and Tracking with Visual Odometry and Automatic Vehicle Control

  • 何 佳瑋

Student thesis: Master's Thesis


This Master's Thesis presents a robot designed for guiding and tracking people from one point to another using only a Kinect as sensor Using Speeded-Up Robust Features feature extraction keypoints are retrieved from the images and stored using the Bag-of-Word model A simultaneous localization and mapping program then performs loop closure to create maps in which the vehicle can localize itself Based on a reference path and the vehicle dynamic model a Model Predictive Control algorithm is used to produce steering angle and velocity inputs designed to optimize a track following trajectory The cost function takes into consideration input variation to provide a smoother trajectory for the user to follow Finally depth clustering and bin summing performed on the point clouds to extract regions of interest ready for upper-body detection The person detections are then transmitted to a multi-hypothesis tracker that determines overlap-free people tracks over time The vehicle control inputs are then transmitted to the vehicle controller based on the detection and tracking of the user to be guided This Master's Thesis thus provides a real-time operating framework able to automatically guide its user along a smooth track while monitoring his presence
Date of Award2018 Jul 13
Original languageEnglish
SupervisorJiun-Haur Tarn (Supervisor)

Cite this