Real time histogram computation in kernel based tracking system

This paper presents an FPGA based embedded hard core IP design for real time moving object tracking system. A well known Kernel based mean shift algorithm is used for tracking a moving object in a real time scene. Implementing these systems in real time applications when there are variant target model and background illumination changes become challenging due to high computational cost and algorithm complexity. We propose the design of some application specific hardcore IPs and to identify reconfigurable hardware blocks. The hardware blocks are designed for bin calculation, kernel derivative, and weighted histogram computation functions of mean shift algorithm. These blocks are used as hardware IPs in embedded architecture and implemented on Spartan-6 IVK FPGA board. The main idea of this work is to accelerate the performance of object detection system by partitioning the algorithm between dedicated hardware and software. The hardware cost of design IPs are compared with some other tracking systems. However, identification of reconfigurable blocks is also part of this work to improve overall performance of the system.