Research On Depth Map Enhancement Algorithm Based On TOF Camera

In recent years, TOF depth cameras have more and more widely used in 3D reconstruction, human-computer interaction, augmented reality and other fields. TOF depth camera have incomparable advantages in high precision, but traditional TOF cameras have disadvantages, for example, the slowly speed, not easy operation and high cost. With the born of Microsoft Kinect, the new cheap and efficient TOF depth camera already have a lot of academic research and wide applications in human-computer interaction, video games, computer graphics and computer vision and in other fields. This type of depth camera has many incomparable advantages compared with the other camera, no matter in its small size, easy operation and low cost. Integrated the advantages, it is widely used in 3D scanning, and it can be applied to the entertainment industry and consumer electronics, historical sites protection, medical images and surgical planning, robot(interaction and navigation) and detection and reverse engineering and so on. But it also has disadvantages. Because of the existence of the noise, the depth image captured by the device is of low-resolution. Therefore, how to obtain fine, smooth and high quality depth maps through the rough and even incomplete information is a difficult problem. With the support of the Shandong province science and technology development plan "Reach and development for the portable 3D body measurement system based on depth camera ", this paper focus on the enhancement of the depth map. In order to get the same resolution as RGB images, using the bi-cubic interpolation algorithm to up-sample a low-resolution depth map, thus a high-resolution depth map can begotten. Because of the noise problems, a new optimization framework is proposed to enhance the depth map.The thesis put forward a new weighting mechanism, to combine the depth map with high resolution RGB image in order to produce a new depth map with more smoothing details.In terms of eh confidence weighting term, not only color similarity and space similarity is used, but also the color gradient is introduced to avoid overflow caused in the process of reconstruction of depth. Because of the artifacts problems, the Gaussian weighted median filter is introduced to eliminate the artifacts and preserve the edges.The algorithm is applied to the processing of the depth maps, and compared to the existing image enhancement algorithm. Plenty of experiments show that the proposed algorithm can do better in preserving edges, eliminating noises, and has the minimum root mean square error. This method is practical and effective.