Fast ROI-based HEVC coding for surveillance videos

In recent years, surveillance videos are playing an important role in modern security systems. As a state-of-the-art video coding standard, High Efficiency Video Coding (HEVC) has been deployed in many surveillance cameras. Nevertheless, the request of real-time and high subjective quality coding still poses the great challenging problem of efficient representation and compression for high complexity HEVC. In this paper, we present a fast Region-of-Interest (ROI) based HEVC coding system for surveillance videos. First of all, the automatic segmentation is conducted to generate the ROI mask. The mask is then used in the quadtree-based block partitioning algorithm and the rate control algorithm so as to reduce the computational costs on non-ROI regions. In particular, we compress the non-ROI regions with a two-step method: fast prediction mode selection and fast coding unit(CU) depth level (size) selection. The ROI-based rate control method is performed as well. The bit allocation between different regions depends on the proportion of ROI region in the whole picture. The proposed system can significantly reduce the computational costs with negligible loss in terms of PSNR at a given bitrate, while the visual quality of ROI can be much improved.