In many real-world video analysis systems , the available resources are constrained, which limits the image resolution. However, the low computational complexity and fast response for low-resolution images still make them attractive for computer vision applications. This work presents a new model that uses a least-mean-square scheme to train the mask operation for low-resolution images. This efficient and real-time method, which uses an adaptive least-mean-square scheme (ALMSS), uses the training mask to detect moving objects on resource-limited systems. The detection of moving objects is a basic and important task in video surveillance systems, which affects the results of any post-processing, such as object classification, object identification and the description of object behaviors. However, the detection of moving objects in a real environment is a difficult task because of noise issues, such as fake motion or noise. The ALMSS method effectively reduces computational cost for both fake motion environment. The experiments using real scenes indicate that the proposed ALMSS method is effective in the real-time detection of moving objects. This method can be implemented in hardware for high-resolution applications, such as full-HD images. A prototype VLSI circuit is designed and simulated using a TSMC 0.18 μm 1P6M process.
Journal of Real-Time Image Processing 13(2), p.311–325