P000203R1
Kinect-like sensor depth image enhancement using 2D/3D feature guidance filtering
Cheng Chen (University of Siegen, Germany)
张勇 (北京工业大学多媒体实验室)
Wennan Chai (University of Siegen, Germany)
*郭铁柱 (北京工业大学多媒体实验室)
Structured light based sensors have attracted intensive research attentions in the fields of mobile robotics, computer vision and pattern recognition due to its super low cost and weight. But the depth image is known to be noisy, poor accuracy and incomplete. The application field of such depth sensor is therefore limited. In this paper, a 2D edge feature guided image filtering algorithm is proposed to enhance the depth measurement. An edge guidance image is used as the guided image, which is calculated based on a probabilistic feature correspondence window searching algorithm between 2D/3D features. Experimental result shows that the quality of the depth image around the edges is significantly improved if the correct correspondence has been found.