A Two-stage H.264 based Video Compression Method for Automotive Cameras
Yiting Wang, Pak Hung Chan, Valentina Donzella
Abstract
With the development of automated vehicles and advanced driver assistance systems, the compression of the large amount of data generated by the vehicle camera sensors becomes a necessary processing step to improve the automated driving system efficiency. H.264 is a widely adopted video compression scheme, and it has been designed for human vision. Rate control in H.264 uses fixed quantisation parameter, however, this process can lead to fluctuation in different regions of the image quality of each frame. In this paper, we propose a two-stage H.264 based video compression framework, named “Two Stage Compression (TSC)”, to compress the automotive camera videos with different values of compression rate in different regions of each frame. In the first stage, each frame will be divided into the region-of-interest and the region-out-of-interest. In the second stage, different compression ratios will be applied based on the importance of the region. The experimental results show that under the same overall compression ratio, our proposed TSC increments the semantic-aware PSNR by 3.213 dB compared to uniform H.264 compression. Our method is also compared to uniform H.264 compression using a segmentation algorithm, with an improvement of 1.77% in mIOU, the average Intersection over Union.