An unmanned aerial vehicle (UAV) with a plurality of cameras uses the cameras capture images, defines an array of regions of each image, and applies an image classifier to a region, classifying the region as either containing an obstacle to be avoided or as a non-threat (fig. 1). If a region is classified as containing an object a second and third classifier may be applied to further classify the type of object, and the outputs of the classifiers may be fused into a final classification by forming a weighted average. The classifiers may estimate the relative locations of any identified objects. The cameras 5A,5B define a collective field of view (FoV) larger than the FoV 405A,405B of any individual camera and the individual camera FoVs overlap so that every image region of every captured image corresponds to at least one image region of another captured image. The regions chosen for classification may be selected based on feature analysis or saliency analysis of each region. The cameras may be visible light or infra-red cameras. A method of detecting an obstacle to be avoided by a UAV is also claimed.
Vision system and method for unmanned aerial vehicles
17.08.2022
Patent
Elektronische Ressource
Englisch
Vision system and method for unmanned aerial vehicles
Europäisches Patentamt | 2022
|Vision based calibration system for unmanned aerial vehicles
Europäisches Patentamt | 2018
|