A Two-stage H.264 based Video Compression Method for Automotive Cameras

Semantic-Aware Video Compression for Automotive Cameras

“…Their effectiveness, however, has not been optimised for usage in video transmission specifically for AV and AAD functions. Some concepts related to rate control in the above mentioned standards are covered in [12], [16]. Recently, newer codecs (e.g.…”

“…Chan et al have demonstrated that object detection based on compression-tuned neural networks can have improved performance in relation to neural networks trained on uncompressed data (in terms of average precision and recall) when lossy compressed video data are transmitted over the vehicle data communication networks [10], [11]. Recent work by Wang et al has proposed some preliminary results on H.264 compression of a user-defined region-of-interest using video data [12]. However, the amount of work on AAD-specific compression is very limited, and additional research is necessary to fully explore the connection between video compression and machine learning (ML)-based perception for automated driving.…”

“…Specifically, as a part of this work, ML-based pre-compression semantic segmentation is used to identify ROI and enable sufficient flexibility to cope with the complexity and variability one can have between consecutive frames in automotive [13]. Following the settings from [12], ROI is identified to include the following labels: road, object, human, parking, vehicle, unlabeled, dynamic, ground; construction/structures, nature, sky are regarded as non-ROI; compression to the different regions is applied using H.264 and H.265 compliant codecs. The contributions of this work are as follows:…”

Optimising Faster R-CNN Training to Enable Video Camera Compression for Assisted and Automated Driving Systems

“…object detection, segmentation, tacking, etc.) and if possible to optimise their interaction [10], [11]; the other one is related to partially or completely(end-to-end) substituting traditional algorithms with machine learning (i.e. deep neural networks) methods and compare the performance of these NN techniques with traditional compression methods.…”

“…in excess of 40 Gb/s) for higher levels of autonomy (levels 3-5), current wired vehicle data networks are inadequate to reliably transmit the required data amount [3]- [5]. With the increased demand of automotive cameras providing high resolution (8)(9)(10)(11)(12) and the required high dynamic range (HDR) to cope with the luminosity variations when driving (e.g. bright sun in front of the sensor when travelling in a dark tunnel), cameras can significantly contribute to the amount of generated data by the sensor suite; moreover multiple cameras are required to provide 360°coverage of a vehicle's surroundings.…”