FPGA Implementation of Object-Based Real-Time Object Tracking Architecture

1. Introduction Detecting moving objects in a video sequence and object tracking are indispensable technologies for surveillance and recognition systems. Various solutions have already been proposed for video object tracking (e.g. background difference based method [1], optical flow based method, and model-based method). However, each method has some short comings, for example, although the background subtraction method can be realized with relatively simple operations, it cannot be applied to the case of a moving camera. Also the optical flow based method needs complex calculations, so that real-time processing becomes difficult. To overcome these problems, we have proposed a multi-object tracking algorithm based on image segmentation and simple object-feature matching [2]. We have confirmed the effectiveness of the proposed method which takes advantage of the spatial object information derived from image segmentation results for difficult cases such as rotating objects, partial occlusion between objects, or a moving camera. In this paper, we introduce an FPGA implementation architecture for the proposed algorithm and report the verification of its effectiveness with our developed FPGA-based object tracking prototype system.