Multi target tracking and detection using LIDAR and velocity obstacles for real time definition of collision zones

• Autor(a) principal: Costa, Rui Pedro Leite Carvalho
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/31317
• topic: LIDAR; Obstacle detection; Collision detection; Velocity obstacles; ROS framework

The implementation of autonomous vehicles on public roads faces many challenges, being the most important ensuring the safety for everyone within that environment. In order for these unmanned machines to be applied in our every-day lives, they first need to be able to correctly identify all possible threats, to then act accordingly. As part of the Atlas project, this dissertation main goal is to determine these threats based on the velocity and dimensions of the detected obstacles. The detection is made using two 2D laser sensors SICK LMS151. These sensors, strategically placed in the front bumper of the ATLASCAR2, allow to perceive its surrounding in an almost 360° (except in the back near the trunk). A detection and tracking algorithm, previously developed at Universidade de Aveiro, was used and refurbished to determine the obstacles’ velocity used to verify if the ATLASCAR2 is in a collision path. Initially, the detection and tracking algorithm was tested using a static LIDAR to retrieved data, resulting in good performances. However the deployment in dynamic environments led to a study of the influence of the ATLASCAR2 ego motion, specifically while it is turning, on the perceived obstacle velocity. The LIDAR’s moving coordinate frame imposes an apparent velocity on the obstacles making the tracking algorithm’s data not as reliable. From this study resulted also a ROS application which allows to visualize the short term path of the ATLASCAR2