Jonathan Brookshire

The Tactical Teams project involved coordinating humans and robots in small, dismounted teams. The first task, thus, was to enable the robot to move with the team without requiring the full attention of the operator (i.e., without requiring teleoperation). To do this, we developed a pedestrian detection algorithm. Fundamentally based on the Dalal/Triggs person detection technique, the algorithm uses histograms of oriented gradients of image regions as features. A machine learning algorithm isolates distinctive human/non-human features from a set of training images. The result is a detection algorithm that runs at ~14 FPS, fast enough to run in real-time on the mobile robot.