Research Of Visual Navigation Approach For Indoor Mobile Robot

ABSTRACT:Smart indoor mobile robots have been the research hotspot in robotics industry. In the environment where human cannot get in, such as inside transformer substation and the stricken nuclear station, indoor robot is needed to accomplish some specific maintenance and rescue tasks. Indoor navigation of mobile robot is the basic and key technology for the robot to accomplish the complex tasks independently. In this thesis, we focus on the algorithms of information collecting and understanding in typical indoor environment.We established a multi-language visual platform via dynamic link library (DLL), using Lab VIEW as the framework, Bumblebee xb3stereo camera from PointGrey Inc. as the visual sensor, FlyCapture for setting up the camera and visual data acquisition, Triclops for image calibration, OpenCV function library for visual data processing.In the indoor corridor environment, a template of four edge lines for the corridor was set up and an iterative algorithm was proposed to track the vanishing point and the edge lines. We deduced the iteration evolving equation of the template based on the Least Square approach. Then we locally sampled along the template and got the edge information nearby, which can drive the template evolving to detect and track the corridor edge lines iteratively. In the more complex environment, a robust algorithm was proposed to track the vanishing point and calculate the camera angle, based on the Hough Transform. We improved the gradient Hough Transform, introduced the reference point of Hough Transformation, and directly calculated the coordinate of the vanishing point in the Hough parameter space by fitting the trigonometric curve, based on the weighted Least Square Fitting and according to the mapping pattern of intersecting lines to the parameter space. By searching the peak value in the parameter space, the real-time camera angle can be calculated.In the complex indoor environment, with the help of line laser as the structure light, we reconstructed the three dimension information using the binocular camera. By introducing the structure light into the system, calculation of feature point matching became more precise in order to get the high-resolution and high-precision indoor three dimension depth map, which can offer information for the robot to understand and environment and make decisions.