MOVING OBJECT DETECTION USING DISCRETE COSINE TRANSFORM BASED BACKGROUND SUBTRACTION

This article addresses the problem of visual detection of moving objects in bad weather conditions. A multi-frame semantic information-based background subtraction scheme is proposed here. The camera is static, hence, the viewpoint is assumed to be fixed. It exploits the spatial as well as temporal neighbourhood at the pixel level by using motion parameters to detect the position of the objects in the field of view. A local attribute map is generated by analyzing the Discrete Cosine Transform coefficients. Further, a spatio-contextual framework is used to obtain the global attributes. Then, the local and global attributes are combined using an entropy-based fusion strategy to get the moving objects in bad weather sequences. The efficacy of the scheme is evaluated using the benchmark bad-weather dataset of CDNet. To check the stand of the proposal among seven recent state-of-the-art schemes, qualitative as well as quantitative analyses are carried out. The results are found to be encouraging.