Integrated Image Processing Unit for Automotive Cameras

The sponsor owns and operates autonomous trucks whose behavior is determined by a combination of software and hardware, and which use a single graphical processing unit, or GPU, for all eight on-board cameras. This design requires 1600 watts to process an image set and represents a single point of failure.The new design uses multiple redundant GPUs designed to handle up to three cameras. The GPUs provide the server on the vehicle with data about distances to objects and types of objects, as well as images that have been enhanced with respect to their brightness, gamma, contrast, boundaries, and denoising variables. The alleviation of computational power from a central unit allows for faster image processing before the image sets and data are sent to the on-board server responsible for vehicle control. This means the server can focus on vehicle control rather than image processing. If a failure occurs in one of these GPUs, the vehicle can continue moving safely. The system requires only 60 watts to complete the analysis of an image set and total power usage does not exceed 300 watts.