LAUSR

The rapid development in the field of computer vision has encouraged researchers to develop vision systems for moving object detection in embedded surveillance applications. The model requires a fast processing algorithm with minimum complexity, which also consumes less memory for computation. This paper proposes a fast moving object detection algorithm to aid foreground segmentation in embedded applications. Dimensionality based Grouping Particle Swarm Optimization-Motion Saliency Map, a variant of the PSO framework combined with saliency map technique is proposed to achieve tighter object detection. The presented technique utilizes the concept of saliency map followed by shadow removal, partial occlusion detection, and Local Difference Pattern based removed object detection. Dimensionality based Grouping Particle Swarm Optimization-Saliency Map of the background model and Dimensionality based Grouping Particle Swarm Optimization-Saliency Map of the incoming frame are used to construct Motion Saliency Map. The proposed model produces a tighter object region compared to the existing naive saliency map based methods. An enhanced texture feature extraction strategy, named as Local Difference Pattern is proposed for removed object detection. This presented moving object detection method is simple and efficient. It consumes less memory for computation. Hence, the algorithm is suitable for embedded surveillance applications. The experimental results show the effectiveness of the proposed method in terms of average processing time in addition to qualitative, and quantitative analyses.