Regresar

Gate recognition and reconstruction for DARPA robotics challenge using Bayesian classifier optimized by mahalanobis distance

Abstract:

Darpa Robotics Challenge contest has different goals that a humanoid robot must achieve in order to reach the next contest stages. The robot has different sensors, such as cameras, gps, etc. In this paper we do not deal with goals but just how to detect the gates based on the mono camera images. We describe how to recognize the posts of the goal based on a simple threshold algorithm. We have used these data for implementing a Bayesian classifier optimized by the Mahalanobis distance for classifying the pixels by color because each post has different color. Then, a Hough lines detector has been used for detecting the lines in the post. Thus, posts are represented by lines. Once detected the posts, the Euclidean distance is used for matching the corner posts with their corresponding pair. Thereby, gate is represented by three interconnected lines. Finally, we have calculated a line from the robot to the center of the gate in order to give the robot a path for arriving to the goal. As an additional work we have detected the pixels of the gate number which is in the gates corner. © 2013 IEEE.