Fast Euclidean Cluster Extraction Using GPUs

Abstract

Clustering is the task of dividing an input dataset into groups of objects based on their similarity. This process is frequently required in many applications. However, it is computationally expensive when running on traditional CPUs due to the large number of connections and objects the system needs to inspect. In this paper, we investigate the use of NVIDIA graphics processing units and their programming platform CUDA in the acceleration of the Euclidean clustering (EC) process in autonomous driving systems. We propose GPU-accelerated algorithms for the EC problem on point cloud datasets, optimization strategies, and discuss implementation issues of each method. Our experiments show that our solution outperforms the CPU algorithm with speedup rates up to 87X on real-world datasets.

FEC: Fast Euclidean Clustering for Point Cloud Segmentation

“…GPU vs. CPU Conventional segmentation methods depend on the CPU's processing speed to run computations in a sequential fashion. Buys and Rusu [39] provide GPU-based version of EC [12] in the PCL [40] library and further extended by Nguyen et al [41] who achieve 10 times speedup than the CPU-based EC.…”

Resilient Navigation and Path Planning System for High-speed Autonomous Race Car

“…3(c). We implement a CUDA-based Euclidean distance clustering algorithm [15], [16] to find cluster C j [17]:…”