Detection of the navigable road limits by analysis of the accumulated point cloud density

• Autor(a) principal: Rato, Daniela Ferreira Pinto Dias
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/28205
• topic: LIDAR; Perception; Curb Detection; Road Limits; AtlasCar2; Point Cloud; Density; ROS; Gradient; Edge Detection

As part of the Atlas project, this dissertation aims to identify the navigable limits of the road by analyzing the density of accumulated point clouds, obtained through laser readings from a SICK LD-MRS sensor. This sensor, installed in front of the AtlasCar2, has the purpose of identifying obstacles at road level and from it the creation of occupation grids that delimit the navigable space of the vehicle is proposed. First, the point cloud density is converted into an occupancy density grid, normalized in each frame in relation to the maximum density. Edge detection algorithms and gradient filters are subsequently applied to the density grid, in order to detect patterns that match sudden changes in density, both positive and negative. To these grids are applied thresholds in order to remove irrelevant information. Finally, a methodology for quantitative evaluation of algorithms was also developed, using KML files to define road boundaries and, relying on the accuracy of the GPS data obtained, comparing the actual navigable space with the one obtained by the methodology for detection of road boundaries and thus evaluating the performance of the work developed. In this work, the results of the different algorithms are presented, as well as several tests taking into account the influence of grid resolution, car speed, among others. In general, the work developed meets the initially proposed objectives, being able to detect both positive and negative obstacles and being minimally robust to speed and road conditions.