Correlation-Based Template Tracking Of Moving Object

Abstract

This paper presents the correlation-based motion estimation technique for the 3D displacement of objects. Two high-speed cameras are configured as a stereovision system and synchronized in real-time. Finger and hand motions are captured in form of digital images at 1500 fps and 2000 fps respectively. A complete motion acquisition system is calibrated to determine the intrinsic and extrinsic parameters which were later used in the correlation algorithm. The grayscale image frames acquired from the cameras are correlated using square templates of 10x10 pixels created from the reference image. The finger and hand motion are discussed with varying camera speed as a measure of brightness inconsistency. The observations in the correlation coefficient indicate that the proposed algorithm is efficient up to 20 and 50 templates for the finger and hand motion cases respectively. The correlation coefficient for finger motion was increased to 0.987 and 0.972 for the left and right cameras, respectively, while the correlation coefficient for hand motion was 0.924 and 0.898. The proposed algorithm is developed in MATLAB and validated by tracing the sinusoidal motion of a solid rectangular element from the image correlation technique and an accelerometer sensor mounted over the block.