Neighborhood based codebook model for moving object segmentation
Priyadarshi Kanungo, A. Narayan, Prabodh Kumar Sahoo, Satyasis Mishra · 2017
For real time surveillance system, a model that can segment foreground from a scene under limited memory having dynamic background and illumination variation is always desired. One such model to address these problems is codebook model. But it consumes a lot of memory as it constructs the codebook for each and every pixel in the image considering that each pixel is unique in nature. Thus it stores redundant values in the codebook resulting in lager memory use. Hence a neighborhood based codebook model (NCBM) using windowing technique is proposed in this work where the codebook is constructed considering a group of pixels corresponding to a window rather than a single pixel at a time. Experimental result shows the least use of memory and better foreground detection compared to basic codebook model (BCM) and adaptive codebook model (ACM). We have used a measure called average codeword per pixel (ACP) as performance measure to estimate the memory utilization. Similarly false positive (FP) and precision (P) is used to estimate the detection efficiency.