The Impact of Lossy Compression on Raw Bayer Images in the Automotive Environment

Digital cameras often use a color filter array, such as the Bayer array, to capture raw color data through single-color wavelength sensors, with interpolation generating full-color (RGB) images. In bandwidth-limited systems like vehicle communications networks, as sensor resolutions continue to increase this creates a need to process and compress data at the sensor level before transmission or storage. Previous work has shown that it is also possible to use raw Bayer data in an automotive machine vision pipeline, however, where bandwidth is at a premium, raw images still incur high transmission costs, making compression essential here as well. This study evaluates how lossy compression affects Bayer images in terms of machine vision performance and image quality, comparing them to full-color (RGB) images with the same compression methods applied. Results show negligible changes in image quality and object detection performance at most compression ratios above ≈ 0.3 Bpp. At very high compression ratios, both image quality and detection performance degrade quickly with higher compression. These findings suggest that Bayer images can be compressed to high ratios with traditional compression methods without significantly compromising object detection performance, facilitating large-scale Bayer image generation suitable for use in automotive machine vision.