Development of Lidar/Camera/Radar-Based Tracking Algorithm and Its Application to Autonomous Driving Vehicles

Abstract

Autonomous driving vehicle is the technology that will appear in daily life in the future By 2025 the world will see millions of autonomous vehicles on the road and its safety is undoubtedly the most important topic in this field In this thesis a multi-sensor data fusion system for multiple computers autonomous vehicle is developed Environ-ment perception is the foundation for ensuring safety so accurate obstacles detection and tracking are critical for autonomous driving Autonomous driving system contains vehi-cle sensors communication equipment and on-board computers To improve system performance this thesis proposes a parallel computing mechanism on two computers and an integration architecture for sensor data processing and decision-making Multiple sensors including camera lidar and radar provide various data for this goal Raw data preprocessing data format definition and fusion algorithm development are integrated in the NCKU autonomous vehicle system based on Robot Operating System (ROS) which provides a good modular development environment Camera has good classification ability; Lidar performs well in bad lighting envi-ronment; Radar provides long-range coverage and good velocity information Deep Learning and Kalman Filter are used to implement detection-level and tracking-level fu-sion algorithm Sensor fusion mechanism of autonomous vehicle is experimented on the campus road with vehicles and pedestrians In the thesis pedestrians tracking perfor-mance is in focus In order to increase safety and decrease road accidents of autonomous vehicles tests in a controlled environment and/or open fields are essential In the future we will make efforts to adapt the system from suiting the campus road to suiting the ur-ban road

Date of Award	2019
Original language	English
Supervisor	Jyh-Chin Juang (Supervisor)