Integrating Automated Object Detection into Mapping in USARSim

Abstract

Object recognition is a well studied field of computer vision, and has been applied with success to a variety of robotics applications. However, little research has been done towards applying pattern recognition techniques to robotic search and rescue. This paper describes the development of an object recognition system for robotic search and rescue within the USARSim simulator, based on the algorithm of Viola and Jones. After an introduction to the specifics of the object recognition method used, we give a general overview of how we integrate our real-time object recognition into our controller software. Work so far has focused on victims’ heads (frontal and profile views) as well as common objects such as chairs and plants. We compare the results of our detection system with those of USARSim’s existing simulated victim sensor, and discuss the relevance to real search and rescue robot systems.