Target following camera system based on real-time recognition and tracking

Abstract

A real-time moving target following camera system is presented in this study. The motion of the camera is controlled based on the real-time recognition and tracking of the target object. Scale Invariant Feature Transform (SIFT) based recognition system and Kanade-Lucas-Tomasi (KLT) tracker based tracking system is presented to recognize and track the moving target. SIFT algorithm is slow but efficient in recognizing the objects even though they undergone some affine transformations. KLT tracker algorithm is simple and has reduced computations, hence improves the tracking performance. The analysis is performed in hardware which consists of a camera mounted on a two servo motor setup, one for pan and other for tilt, and an Arduino board capable of handling the movement of two servo motors. As there is hardware implementation, a computationally simplified technique is employed. Since both SIFT and KLT tracker are feature based techniques, we pass the features extracted by SIFT to KLT tracker for simplifying the process. The recognition and tracking tasks are performed in PC and the PWM signals are generated accordingly and sent to servo motors through Arduino. The proposed algorithm is able to track objects even in its absence for a certain while.