Real-time video freezing detection for 4K UHD videos

摘要

Video frame freezing is a common artifact which can occur during video content delivery due to errors in the video coding process, video transmission, storage or reproduction. This artifact can significantly decrease the end-user Quality of Experience. Therefore, accurate video frame freezing detection is of great importance for different parties involved in video content delivery. In this paper, a new Real-Time no-reference Freezing Detection Algorithm, called the RTFDA, is proposed. As newly video frames are acquired, the RTFDA performs comparison of the current video frame with the corresponding previous one to detect whether video freezing is occurring. The comparison is made by calculating the corresponding subsampled video frames and their pixel-by-pixel absolute difference comparison. The benefit of such approach is twofold: the influence of noise in freezing frames on frame comparison is significantly reduced as well as computational complexity of frame comparison. The RTFDA has a high detection rate with a very low rate of both false-positive and false-negative detections, outperforming four freezing detection algorithms on four different video databases. The proposed implementation on an x86-64 platform achieves real-time performance on 4K Ultra High Definition (UHD) videos by processing 216 frames per second (fps). Apart from that, FPGA implementation is proposed, which has efficient FPGA resource utilization.