Multisensor integrated stair recognition and parameters measurement system for dynamic stair climbing robots

In this paper, we develop an online real-time system using Kinect and inertial measurement unit (IMU) to recognize and measure stairs for stair climbing tasks of erect mobile robots. This system combines the ideas of statistical characteristics of 3D point cloud data and adjustable geometric models for stair recognition tasks. Through this system, we not only verify the advantages of statistics methods in 3D environment perception process, but also realize the idea of using adjustable geometric models to fit 3D perception data correctly and efficiently. Because our stair recognition system keeps good quality results (average errors within ±0.4 cm) and less time consumption (less than 0.1 second per frame) even under dynamic situations, it is expected that it can help mobile robots to achieve online dynamic stair climbing tasks.