Binary histogram based split/merge object detection using FPGAs

Tracking of objects using colour histograms has proven successful in various visual surveillance systems. Such systems rely heavily on similarity matrices to compare the appearance of targets in successive frames. The computational cost of the similarity matrix is increased if proximate objects merge into a single object or a single object fragments into two or more parts. This paper presents a method of reducing this computational cost with the use of a reconfigurable computing architecture. Colour histogram data of moving targets are used to generate binary signatures for the detection of merged or fragmented objects. The main contribution in this paper is how binary histogram data is generated and used to detect split/merge object with the use of logical operations native to the hardware architecture used for its implementation. The results show a 10 fold improvement in processing speed over the microprocessor based implementation, and that it is also capable of detecting split/merge objects efficiently.