engleski

Scene Reconstruction and Free Space Representation for Biped Walking Robots

This thesis presents a scene interpretation strategy designed for application in the vision-based guidance of biped walking robots in indoor environments. Perception of the environment is achieved by using a stereo camera pair together with robot sensor data. The presented scene reconstruction strategy allows interpretation of poorly textured indoor scenes as well as precise determination of obstacle size and the boundaries of walking surfaces. The high accuracy of environment information, critical when stepping over obstacles or climbing stairs, is achieved by combining straight line segments detected in the images acquired by one of the cameras with depth information provided by a correlation-based stereo vision method. The 3D environment model obtained by the applied scene reconstruction methods is used as the basis for determining the free space for biped walking robot. A novel method is presented which creates a graph representation of the free space based on an approximation of the robot by a set of hulls. Their shapes allow efficient determination of feasible paths involving stepping over obstacles and changing the walking level in an indoor scenario. The applicability of the presented methods in the guidance of walking robots is evaluated using camera images and robot sensor data recorded during experiments with a real walking robot.

walking robots; vision-guided robot walking; scene reconstruction; free space representation

nije evidentirano