GPU-accelerated rdp algorithm for data segmentation

Abstract

The Ramer-Douglas-Peucker (RDP) algorithm applies a recursive split-and-merge strategy, which can generate fast, compact and precise data compression for time-critical systems. The use of GPU parallelism accelerates the execution of RDP, but the recursive behavior and the dynamic size of the generated sub-tasks, requires adapting the algorithm to use the GPU resources efficiently. While previous research approaches propose the exploitation of task-based parallelism, our research advocates a general fine-grained solution, which avoids the dynamic and recursive execution of kernels. The segmentation of depth images, a typical application used on autonomous driving, reaches speeds of almost 1000 frames per second for typical workloads using our massively parallel proposal on low-consumption, embedded GPUs. The GPU-accelerated solution is at least an order of magnitude faster than the execution of the same program on multiple CPU cores with similar energy consumption.