Effect of Temporal and Spatial Noise on the Performance of Hardware Oriented Background Extraction Algorithms

Background extraction is a fundamental task present in most computer vision applications such as video surveillance, optical motion capture or multimedia applications. In this paper we explore a particular foreground segmentation method based on the well-known Pixel-based Adaptive Segmenter (PBAS) algorithm, proposing modifications that will ease the hardware implementation. Also, the figures of merit of a focal-plane approach for foreground segmentation are studied through the impact of typical temporal and spatial noise sources present in the processing elements of smart image sensors such as leakage currents from analog memories or fixed pattern noise (FPN) from mismatch.

keywords: Smart Image Sensors, Focal-Plane Processing, Background subtraction, PBAS, hardware implementation.