Development of a Detection and Tracking of Moving Vehicles system for 2D LIDAR sensors

More Info
expand_more

Abstract

The main objective of this thesis was the development and evaluation of a Detection and Tracking of Moving Objects (DATMO) system, that is capable of reliably tracking nearby vehicles from a moving car. The developed system takes in raw 2D LIght Detection And Ranging (LIDAR) measurements as input and detects objects of interest by clustering them with the Adaptive Breakpoint Detector algorithm. The resulting clusters are fitted with oriented bounding boxes, by incorporating the Search-Based Rectangle Fitting algorithm. The tracking part of the system receives, extracted from the rectangles, L-shapes and associates them with already tracked vehicles, using the Global Nearest Neighbor algorithm. However, since LIDAR measures only the distance to surfaces that face the sensor, vehicle appearances change over time. In order to counteract tracking errors that originate from these changes, an L-shape switching algorithm is implemented. The kinematic poses of the tracked vehicles are estimated with two different tracking filters, a Kalman Filter (KF), with a constant velocity model and an Unscented Kalman Filter (UKF), with a Coordinated Turn kinematic model. The dimensions of the detected vehicles are estimated with a constant shape Kalman Filter. The proposed system was evaluated using both simulation and real-world experiments. The real-world experiments were made on the Delft Scaled Vehicle (DSV), an experimental car platform in the form of a 1/10 scale radio controlled car, and the ground truth data were provided by a Motion Capture (MoCap) system. The simulation experiments were made in a highway environment, which was created specifically for the development and testing of this system. Evaluating the experiment results reveals that the developed system can reliably estimate the position, speed, heading angle and dimensions of surrounding vehicles and therefore it can be used in similar research platforms to expand their environment perception capabilities.