| We present a method for navigating an autonomous agent based on the textures present in an environment. Specifically, the autonomous agent in question is that of a robotic lawn mower. If we can successfully differentiate the textures of the cut and uncut lawn surfaces, then we can track the boundary between them and mow in a pattern as a human would. This dissertation covers the problem of detecting different textures present in an image and then using that information to guide an autonomous robot. The system uses the wavelet transform as the basis to perform texture analysis. The wavelet transform extracts meaningful features from the input images by breaking the image into different frequency subbands. Different subbands will isolate different features in the input image. In this way, we can generate a frequency signature of the image. After performing the wavelet transform, we perform a post-processing stage on these resulting features in an attempt to make them more acceptable to our classifier. These processed features are then grouped into vectors and then classified. The result is a clustered image based on texture.; Once we have the image segmented based on the textures present in the image, we then determine the boundary between them by use of a boundary detection algorithm. In this way we can give a robotic lawn mower the ability to track this boundary and mow as a human would mow. While we avoid the actual implementation of this algorithm on a real platform due to the hazardous nature of lawn mowing in general, we do show how this algorithm can be easily adapted to the task of sidewalk tracking. In this alternate task, the robot tracks the boundary on both sides of the sidewalk, giving the robot the ability to follow the sidewalk. In doing so we not only show the adaptability of our algorithm to another task but also show its implementation on a mobile robot platform. |
