Accelerating Adaptive Background Modeling on Low-Power Integrated GPUs
Shoaib Azmat, Linda Wills, Scott T. Wills · 2012
Background modeling is a key initial step in many video surveillance applications. As more and more smart cameras are deployed for surveillance tasks across the globe, an efficient background modeling technique is required that balances accuracy, speed, and power. Due to its high parallel computational characteristics, robust adaptive background modeling has been implemented on GPUs with significant performance improvements over CPUs. However, these implementations are infeasible in embedded applications due to the high power ratings of the targeted general-purpose GPU platforms. We propose implementing a fast, adaptive background modeling algorithm on a low-power integrated GPU, the NVIDIA ION, with thermal design power (TDP) of only 12 watts. This paper focuses on how data and thread-level parallelism is exploited and memory access patterns are optimized to target this algorithm to a low-power GPU. We achieve a frame rate of 100fps on a full resolution VGA (640x480) frame. This is a 6X speed-up compared to a CPU platform of comparable TDP.