关键词：Kinect；相机；目标检测；计算机视觉；数据集
摘 要：With the arrival of the Microsoft Kinect, obtaining depth maps of interior spaces has become remarkably easy. The Kinect is equipped with an 8-bit RGB VGA resolution (640x480 pixel) video camera, and also features an IR-triangulation based depth sensor with reports of accuracy within q(z) = 2:73z2 + 0:74z 0:58[mm], with z the depth in meters. The Kinect's low cost and portability make it an attractive instrument for robotics and mapping. We have witnessed a boon of large datasets originating from such Kinect-style cameras, and an associated development in algorithms for SLAM-like tasks. While the potential for this data is vast, one immediate application is incorporating the depth data into a more robust object detector.The remainder of this report is organized as follows: Part 2 describes previous work done in the area of object recognition and depth data. Part 3 describes the dataset. Part 4 gives our approach to the problem. Part 5 gives results of our method. Part 6 provides closing remarks.