High Energy-efficient Change Detection Method for Video Surveillance with Approximate Computing

Video surveillance is widely used for security monitoring of important infrastructure such as highways and electricity facilities. With the development of new technologies and limitations in transmission bandwidth, some functions need to be deployed on edge devices, increasing the demand for computational power. As a typical fault-tolerant application, video surveillance can be designed to be lightweight using approximate computing techniques. This paper focuses on the fundamental function of change detection in video surveillance and presents an approximate hardware implementation of the change detection based on principal component analysis (PCA) and K-means algorithm. Experimental results show that the approximate change detection method produces results with extremely small differences compared to precise calculations, while reducing the hardware area overhead of the core operator by over 40%.