Robot Vision

Robot vision refers to providing robots with the capabilities of visual recognition, thus enabling them to ``see'''' what is ``in'''' an image. This project involves the construction of a system to extract the structure from a corridor scene using images taken from a camera. The corridor structure is determined from a camera image. Component tasks performed on the image are pre-processing, line detection and construction of a corridor map. Pre-processing is acheived using Sobel edge detection and thresholding, providing the edges of high contrast. Line detection is implemented using the UpWrite, providing a set of lines from the image. Using the set of lines found in the image, the structure of the corridor is determined using vanishing points and features within this corridor are located using the corridor description. A map of the corridor is the output of the system, representing the location of the corridor with respect to the robot. This process has proven to be effective in describing the corridor structure and locating various features with accuracy dependent upon locating the correct vanishing point to describe the floor. The system provides an overall speed of 3 frames per second on a $320 \times 240$ image and there is good potential for increasing this rate. This system has practical applications in robot navigation, using visual input.