Outdoor human detection system for mobile robot navigation

Development of a robust outdoor navigation system for a mobile robot is a challenging task. A robust navigation system is expected to provide means for the robot to recover in the case of a navigational failure. This project develops a vision-based human detection system that can be used as a part of the outdoor navigation system of a mobile robot. The system can detect moving or stationary humans in real-time so that the robot can interact with them to get directions if needed. The detection system has two (2) subsystems. The first subsystem assumes the robot is stationary (e.g., stopped). It detects non-interruptive motion and then categorizes moving objects as humans or cars using statistical background modeling technique. The second subsystem uses Cascaded Boosted Classifiers with Haar-like features and it does not assume that the robot is stationary. A large number of negative (10,000 images) and positive (5000) data sets collected outdoor are used to train the classifier. Both of the subsystems have been successfully implemented and tested using a Pioneer 3-AT mobile robot and a system performance analysis has been provided