Monocular Rock Detection

The purpose of this algorithm is the detection of rocks within the rover’s field of view and is based on a supervised learning process using image data. Rocks are detected by finding changes in intensity and size from input examples and learning a binary classification for the pixels. A linear Support Vector Machine (SVM) classifier is used to locate rock features in image data by running a binary classification of pixels. A set of rock features is subsequently interconnected through a standard breadth-first search algorithm. This process offers a mask to the input image with rock candidates. Each individual group of candidate pixels from the rock masks is checked for correspondence to a rock. This procedure is realised through heuristic methods that operate on the shape and the texture of the candidate rock’s image region. The generated data product is a list of rocks from a candidate rock mask as shown in the example images: