Lossless compression of RAW image data on the FPGA

Abstract

The CoSenS research group at UHasselt focusses on the implementation of multi-camera systems. These systems generate a large amount of video data. One central processing system calculates all data and becomes heavily loaded. The solution is data compression. To maintain good cohesion visual clues among images the method must be lossless. First, this master thesis evaluates and compares various compression methods on RAW and RGB data. This is followed up by the implementation of the most suitable method. The compression method used must reach a compression ratio of 60% and a processing speed of more than 40 Mb/s.

The research revolves around compression methods for RAW and RGB data. The filtering of noise is also considered. Criteria for compression methods are: speed, accuracy, loss and compression ratio. The implementation is realized on the Altera DE2-70 board with software Quartus II 8.1 and NIOS II using Verilog. The results are analyzed in Matlab R2016b. The camera TRDB D5M captures the video information.

The implemented method utilizes Rice encoding after data preparation using the GAP algorithm. After calibration to reduce noise, the data is compressed in RAW state. The method achieves a compression ratio of 45% at the speed of 44 Mb/s. In the future, this method can be optimized which can result in a compression ratio of above 50% in a normally lit room