Efficient lossless compression for depth information in traffic scenarios

Modern day automotive features (e.g., in-vehicle augmented reality) require a depth of the environment as the input source. It is important that depth data can be transferred from one processing unit to another in a car. About 10 years ago, Stixel has been introduced as a mid-level representation of depth maps (disparities) which reduces the data volume thereof significantly. Since then, Stixel has been extensively researched and is nowadays a seriously considered solution for series production cars. Nevertheless, even after using a Stixel representation, the depth data can hardly fit into a low- or medium-bandwidth in-vehicle communication system, e.g., via a CAN bus. Hence, the cost-sensitive automotive industry is still seeking new solutions for the transmission of depth information using in-vehicle communication buses. In this paper, we present an efficient lossless compression scheme for Stixels as a potential solution to this problem. Our proposed algorithm removes both spatial and temporal redundancies in Stixels through a combination of predictive modeling and entropy coding. Evaluation shows that it outperforms general purpose compression schemes, e.g., zlib, by more than $$60\%$$ in space savings. More importantly, we prove that using the proposed Stixel compression, depth information could be transmitted through a less expensive CAN bus, whereas a much more expensive FlexRay bus is needed otherwise. We believe that this finding has great relevance for the automotive industry.