Means of detecting an obstacle to be avoided by a UAV. The UAV comprises at least one camera. A method of obstacle detection involves capturing an image 100 with the camera, producing a hazard map 108 corresponding to an array of regions of the capture image by detecting a plurality of points 102 in the image that define the location of image features and/or transforming the image into a saliency image (figure 2). For each region the presence or absence of a hazard is estimated, and a classifier is used to classify regions indicated as containing a possible hazard as either an obstacle to be avoided or non-threat regions. Two or more classifiers may be applied and a fused final classification for a region is determined by forming a weighted average. The hazard map may comprise combined saliency and feature data. The UAV may comprise a plurality of cameras with overlapping fields of view (figure 4). A combined hazard map may be generated covering the combined field of view of all the cameras.
Vision system and method for unmanned aerial vehicles
2022-08-17
Patent
Electronic Resource
English