Sensor-based data compression enables safe autonomous driving Autonomous driving is revolutionizing individual transport, service-oriented mobility concepts and the commercial vehicle sector. The overall system and its components are safeguarded by controllable and reproducible scenario-based tests in real or virtual environments. An in-vehicle edge server concept for demand-oriented data compression is being developed and prototyped for the flexibly adaptable and high-performance processing of the resulting data volumes.

The validation of image-processing ECUs in high-tech applications, such as autonomous driving, is a challenging task with increasing complexity. On the one hand, the number of installed imaging sensors is increasing, on the other hand, their spatial resolution is also increasing and, as a direct consequence, so is the amount of data that is emerging. A result that places new and challenging demands on the systems to be processed. In order to cope with the complex task of high-performance development and safeguarding of applications for autonomous driving, the recording of suitable scenarios in the real vehicle (data logging) is increasingly gaining acceptance. The sensor data stored in this way is then replayed at any later time as required (so-called data replay) in order to test the control units more cost-effectively. In order to make more efficient use of the limited storage capacity of the system used in the vehicle for recording, lossless compression methods - which meet the requirements of the sensor data - are therefore needed.

Within the scope of the transfer project, various state-of-the-art methods for lossless image data compression were investigated and evaluated with respect to their suitability for the image data typical in vehicles. For this purpose, the characteristics of automotive sensor data were first worked out and taken into account in the analysis of the compression methods. Among the methods investigated, a selected algorithm was implemented prototypically in a demonstrator based on the in-vehicle platform AUTERA from the company dSPACE. In order to make the best possible use of the available maximum bandwidth and the storage capacity of the system, the data was compressed as close as possible to the sensor and decompressed only shortly before being fed into the ECU. This criterion, combined with the required real-time capability of the system, necessitated the implementation of a hardware-accelerated (de-)compression solution based on a flexible and high-performance FPGA expansion board of the dSPACE AUTERA in-vehicle platform (cf. figure). The aim was to create a proof-of-concept of intelligent and on-demand data compression for data logging and data replay in the context of autonomous driving.

Requirements, trends and challenges in the field of data logging and data replay of sensor data in the automotive sector were identified and an overview of lossless compression methods for camera data was created. By evaluating with real test data, advantages and disadvantages of common compression methods in the processing of image data from the field of autonomous driving could be shown. With the help of additional, economic evaluation criteria tailored to the target platform, JPEG-LS was selected as the image-based compression method. Its implementation as a flexible FPGA design that can be adapted to different use cases enables it to be used with a high-performance FPGA expansion board from the dSPACE AUTERA in-vehicle platform.