A fast video coding algorithm using data mining for video surveillance

Video surveillance is crucial for various applications, including unmanned aerial vehicle operations, flight safety monitoring, social security management, industrial safety, and criminal detection. The large volume of video data generated in these areas requires efficient processing techniques. However, traditional video compression and encoding methods are often complex and time-consuming, which can hinder the real-time performance needed for effective surveillance systems. To address this challenge, we propose a novel fast coding algorithm optimized for video surveillance applications. Our approach employs frame difference analysis to classify coding units (CUs) into three distinct categories: background CUs (BCs), motion CUs (MCs), and undetermined CUs. For both BCs and MCs, the algorithm examines the probability distribution of potential coding modes and depths, subsequently skipping unlikely combinations to enhance processing efficiency. The remaining candidates are then processed using a decision tree model, which enables accelerated mode and depth selection through early termination. Experimental results show that our method significantly accelerates encoding speed while maintaining almost identical coding efficiency, making it particularly effective for real-time surveillance applications.