On Localization And Object Tracking For IARC Mission 7

Aerial robots have been widely used in aerial photography, transportation, inspec-tion, search and rescue. The localization, navigation and target tracking technologies have attracted much attention in recent years. International Aerial Robotics Competi-tion (IARC) is one of the most challenging competitions in the area of aerial robotics nowadays. It aims at tackling challenges that off-the-shelf technologies are not exis-tence. This thesis is strongly motivated by IARC mission 7. The contributions of this thesis are described as follows.1. A mission-oriented quadrotor system is set up. The system design includes a powerful onboard computer which makes it possible to run high-level task, such as visual odometry, target tracking, obstacle avoidance etc. A manual-autometic switching module is designed, which is switched between remote control mode and onboard computer control mode. This module takes commands from both remoter and onboard computer as input, and sends one of them to low-level flight controller according to the operation mode. The system can be also used in other scenarios like inspection, search and rescue.2, Optical flow is used to estimate the velocity of the quadrotor. We improve the accuracy and smoothness of the velocity by fusing optical flow and EMU data by a Kalman filter. A cascade PID controller is implemented to achieve hover-ing. The partial position is obtained by integrating the velocity over time, while error is also accumulated. Hence, a global localization approach based on the Zhang Zhengyou’s calibration method is proposed. The principle of this method is illustrated, and its feasibility is accordingly analyzed.3. A hybrid system consisting an AR.Drone and an iRobot Create is designed for single moving object tracking and touching under the competition environment. CamShift algorithm is used to recognize the moving iRobot Create and estimate the relative position between AR.Drone and iRobot Create. A path planner and height controller is implemented for the moving object touching. Several experi-ments show the effectiveness of these algorithms.4. Considering there are multiple moving objects in the competition arena, multi-object detector and tracker are designed based on the color characters and the nearest neighbor algorithm. To obtain the moving directions of objects, an orien-tation detector based on the shape of the colored area is proposed. Experimen-tal results show the effectiveness of the algorithms. Videos of experiments are av in Youku channel (http://i.youku.com/u/UMzMwMDg3NTAlMg==/ videos), which gives an intuitive performence of the algorithms in this thesis.