Object-conditioned plane segmentation

Method of identifying a planar surface 4, 6 in an image, comprising: receiving a first scene image and a second scene image; detecting objects 14-20 in the first image; identifying a candidate region that contains no objects or only planar objects in the second image; applying a plane identification algorithm to the candidate region in the second image. The images may be RGB image; RGB-D (depth) images; or point cloud images. The candidate region and object region may not overlap. Determining the candidate region in the second image may comprise removing the detected objects' regions, pixels or point cloud data from the second image. Determining the candidate region in the second image may comprise detecting substantially flat objects (e.g. tables, chairs, shelves) and setting the candidate region to the detected object region. If the scene is too cluttered, the object detection algorithm may identify object bounding boxes 22-26. The candidate region may then be determined by: clustering objects based on the bounding box bases; and removing all points which have a coordinate further than a threshold from the cluster centre. The remaining points form the candidate region to which the plane detection algorithm is applied. The method may be implemented in a robotic appliance.