GPU-Services: Real-Time Processing of 3D Point Clouds for Robotic Systems Using GPUs

Abstract

The GPU-Services project fits into the context of research and development of methods for data processing of three-dimensional sensors data applied to mobile robotics. Such methods are called services on this project, which include 3D point clouds pre-processing algorithms, segmentation of the data, separation and identification of planar zones (ground, roads), and detection of elements of interest (edges, obstacles). Due to the large amount of data to be processed in a short time, these services will use parallel processing elements, using the GPU to perform partial or complete processing of these data. The project aims to provide services for an autonomous car, forcing the services to approach a system for real-time processing, which should complete the whole data processing before the next frame came from the sensors (~10 to 20Hz). The sensor data is structured in the form of a cloud of points, allowing for great parallel processing. However, its major difficulty is the high rate of data received from the sensor (around 700,000 points/sec), and this gives the motivation of this project: to use the full potential of sensor and to efficiently use the parallelism of GPU programming. The GPU services are divided into steps, but always seeking the processing speed given by their intrinsic parallelism: The first step is to organize an environment for parallel processing development in conjunction with the system already being used in our autonomous car, The second step is an intelligent extraction and reorganization of the data provided by the sensor (Velodyne multi-layer laser sensor), The third stage is a pre-segmentation of non-planar data, The fourth stage is performing the segmentation of data received from the previous steps in order to find objects, curbs and ground plane, The fifth stage is to develop a methodology that unite the results of previous steps in order to explore the topology of the environment, i.e. Will aim to structure the results into a topological form (identifying pathways and links between pathways, such as curves and intersections) to assist other projects that focus on vehicle control and autonomous navigation systems.