Vision-Based Detection and Tracking of a Moving Target for a Smart Car

Abstract

In this paper, vision-based detection and tracking of a moving target for a smart car system is proposed and implemented on a smart car loaded with raspberry pi. In target detection, FLANN is used to replace the traditional violence matching method, which significantly increases the quickness and accuracy of feature matching. In our system, ultrasonic data are used to solve scale ambiguity of the monocular camera. To address multiple interfering objects in the initial field of view, we propose the idea of initial hypothesis frame. During the tracking process, the target trajectory is estimated and predicted with the unscented Kalman filter to solve the loss or occlusion of the target. In the controller design, single neuron adaptive PID control is adopted. All these functions and algorithms are implemented on a raspberry pi.