The Advanced Correlation Tracker in Real-Time

Real-time detection, tracking and classification of objects in images are some of the more computationally demanding tasks in computer vision. Tracking systems such as correlation trackers are notorious for computational demands. Real-time performance is not achievable on general purpose microprocessors. This paper details the architectural issues tackled to create a real time realisation of the Advanced Correlation Tracker (ACT) algorithm on an FPGA. The problem was tackled using top-down design and a software driven development cycle. It was found that a strong understanding of the nature and function of the algorithm, as well as the underlying strengths and limitations of the processing architecture, were needed for a successful transition from high-level algorithm to real-time realisation.