Real-Time Number Sign Detection on a Mobile Outdoor Robot

Automatic sign detection is an essential task for robots working in environments built for humans: Signs can serve as landmarks and convey messages for both humans and robots.In this work we try to find the best system for detecting a set of number signs used for marking subgoals in the 2010 “SICK robot day” robotics challenge. The goal of this challenge was to build a robot that autonomously finds and drives to 10 numbered signs in a given order within at most 10 minutes.Hence the system had to run on an on-board processor of a mobile robot and had to provide real-time performance. The mobile robots used for these experiment are custom built based on a remote-controlled monster truck “E-MAXX”by Traxxas. Among several other sensors they contain three cameras: The main camera is a forward-looking AVT Marlin mounted on a pan unit so it can keep a number sign focused while driving around corners. It captures grayscale imagessized 780 x 580 at a rate of 30 Hz.

We additionally employed two Firefly MV cameras by Point Grey Research looking left and right, hoping to find more signs due to the larger total area of view. These two cameras grab grayscale images sized 640 x 480 at a rate of 7.5 Hz. All processing is done on a MiniITX computer featuring a 2.26 GHz Core 2 Duo Mobile CPU mounted on the robot. To read this external content in full, download the paper from the author archives. http://www.researchgate.net/publication/236594024_Robust_Real-Time_Number_Sign_Detection_on_a_Mobile_Outdoor_Robot/file/3deec51829169d055d.pdf