Locating 3D Object Proposals: A Depth-Based Online Approach

Abstract

2D object proposals, quickly detected regions in an image that likely contain\nan object of interest, are an effective approach for improving the\ncomputational efficiency and accuracy of object detection in color images. In\nthis work, we propose a novel online method that generates 3D object proposals\nin a RGB-D video sequence. Our main observation is that depth images provide\nimportant information about the geometry of the scene. Diverging from the\ntraditional goal of 2D object proposals to provide a high recall (lots of 2D\nbounding boxes near potential objects), we aim for precise 3D proposals. We\nleverage on depth information per frame and multi-view scene information to\nobtain accurate 3D object proposals. Using efficient but robust registration\nenables us to combine multiple frames of a scene in near real time and generate\n3D bounding boxes for potential 3D regions of interest. Using standard metrics,\nsuch as Precision-Recall curves and F-measure, we show that the proposed\napproach is significantly more accurate than the current state-of-the-art\ntechniques. Our online approach can be integrated into SLAM based video\nprocessing for quick 3D object localization. Our method takes less than a\nsecond in MATLAB on the UW-RGBD scene dataset on a single thread CPU and thus,\nhas potential to be used in low-power chips in Unmanned Aerial Vehicles (UAVs),\nquadcopters, and drones.\n