3D Medial Axis Distance for hand detection

In this paper, we proposed a Medial Axis Distance (MAD) measure for body part detection with single depth image only. First of all, we extracted the skeleton line of human body based on the detected human body silhouette. Using the space information of pixels on the skeleton line and the silhouette, we determined a human body center point on the skeleton line. Then we proposed the MAD measure which calculated the 3D distance between each pixel of human silhouette to the body center point. The MAD measured the spatial distances from different body parts to the body center, and is capable of distinguishing the limbs and body part on a human body. By doing so, there are two advantages. The proposed MAD measure is capable of detecting human body parts without using any training sample. Moreover, it can work well even under poor illumination. We evaluated the proposed MAD measure for hand detection on the DHG hand gesture database. The experiment results showed that the detection rate of hands reaches 94.2%.