CUDA Based CAMshift Algorithm for Object Tracking Systems

In this paper, we present an image object tracking system for GPGPU based CAMshift algorithm. For image object tracking, we use the parallel CAMshift tracking algorithm based on the HSV color image distribution of detected moving objects. In this, RGB-to-HSV color conversion, image masking such as open and close operation for image morphology, and computing of centroid are executed in parallel. CAMshift algorithm is very efficient for real-time tracking because of its fast and robust performance. In this system, CUDA environment and C++ program are used for image processing and accessing the PTZ protocol and RS-485 communication for controlling the position of PTZ camera in order to arrange the moving object images in the middle part of the monitor screen. This system can be applied to an effective and faster image surveillance system for continuous object tracking in a wider area and real time. Key-Words: CUDA, Object tracking, PTZ camera, CAMshift algorithm, parallel processing