Integral image computation algorithm for GPU-enabled automotive platforms

Nowadays, automotive industry is shapeshifting towards utilization of solutions in which software and algorithms are the dominant concern. Autonomous driving functions require processing-intensive decision algorithms which involve a lot of image manipulation software techniques. CPUs cannot meet such a demand; therefore, hardware acceleration again becomes mandatory in any design. GPUs have become great candidates for image processing in automotive algorithms, due to large number of cores. However, efficient porting of traditional image processing algorithms to GPUs is still a challenge. In this paper, we present an improvement of integral image computation on the GPU in an automotive context. The main benefit of the proposed improvement is reduced computational complexity and consequently speedup of execution time. Therefore, frame rate in an image processing algorithm is increased, which is necessary in modern automotive industry. Furthermore, we address specific challenges such as data dependency, resolution and optimization of arithmetic. Finally, we evaluate the approach utilizing NVIDIA CUDA for a Xavier-based automotive electronic control unit.