New Techniques for Path Planning in Image Space

We propose two new methods for performing local path search directly in the image space of a camera sensor. The first technique utilizes a high-resolution local image space subsystem to determine the farthest point along a global Cartesian path that can be reached along an unobstructed heading, thereby short-cutting as much of the global path as possible. To the best of our knowledge, this is the first image space planning technique that is both suitable for navigation in unstructured environments and does not use a graph-search algorithm in the image space subsystem. The second technique that we propose improves the local image space subsystem of a hierarchical image/Cartesian space planner by determining when it is safe to translate at full speed. Both techniques are implemented on an autonomous robot and experimentally evaluated against a hierarchical top-down Cartesian planner and a previously proposed hierarchical image/Cartesian space planner. Each system is tested three times on three different course layouts. The first proposed method performs comparably to existing systems when evaluated on the criteria of path length and total runtime. The second technique consistently outperforms the other three systems with respect to total runtime.