Impact of the Length of Optical Flow Vectors in Estimating Time-to-Contact an Obstacle

Obstacle detection and avoidance for autonomous aerial vehicles is one of the most researched problem in the field of robotics. Vision-based obstacle detection and avoidance in aerial vehicles has received significant attention recently because of their inherent advantage of low power consumption and less weight. In this work an optical flow based time-to-contact method is used for obstacle avoidance. Though the accuracy of this method is good at close range, it cannot be used when obstacles are far away. This work investigates the reason for its less accurate performance for distant obstacles. It helps to identify a boundary within which the performance of the optical flow based time-to-contact method will be more accurate. Using Robot Operating System(ROS) and Gazebo - a 3D simulator, simulations are performed to estimate the time-to-contact an obstacle for an unmanned aerial vehicle (Parrot AR Drone). The results obtained and the analysis performed helps to identify the reason for the poor performance and suggests a proper boundary within which its performance will be more accurate.