Efficiently Annotating Object Images with Absolute Size Information Using Mobile Devices

摘要: The projection of a real world scenery to planar image sensor inherits the loss information about 3D structure as well absolute dimensions scene. For analysis and object classification tasks, however, size can make results more accurate. Today, creation annotated datasets is effort intensive typically requires measurement equipment not available public contributors. In this paper, we propose an effective annotation method that utilizes camera within smart mobile devices capture missing along with image. approach builds on fact camera, calibrated specific distance, lengths be measured in object’s plane. We use camera’s minimum focus distance calibration adaptive feature matching process for precise computation scale change between two images facilitating measurements larger distances. Eventually, segmented its later analysis. A user study showed humans are able retrieve low variance. proposed facilitates accuracy comparable manual ruler outperforms state-of-the-art methods terms repeatability. Consequently, allows in-situ objects without need additional or artificial reference