Image Based Human Body Pose Estimation And Hand Gesture Recognition

The human body pose estimation and gesture recognition are convenient for human-computer interaction. So for these studies are increasingly hot. This paper studies the single image based human body pose estimation and single depth image based hand gesture recognition.The main contents of this paper are on the following:First, the status quo of human pose estimation is introduced. Pictorial Structure are briefly introduced, along with the Kinect depth camera hardware, as well as data acquisition, conversion, etc. The bilateral filtering method for depth image which is used to reduce the deep image noise is also discussed.Second, an image body pose estimation method based on shape context is proposed. It is difficult to robustly estimate the human pose under the complex background, which is partially due to the lack of good features to separate the foreground human from the complex background. This paper presents a segmentation based method. First, the human is segmented from the background by the semantic segmentation. Then shape context method is adopted to obtain the human pose in the Pictorial Structure. Experimental results show that the proposed method can obtain a better body estimation than the existing method for complex backgrounds.Finally, a depth image based random forest incorporated hand gesture recognition is presented. We use Kinect depth camera to capture depth data of the human body, with bilateral filter to reduce noise of the depth image. The hand is then separated from background so that the depth map of hand is obtained. Consequently, the Random Forest method is used to classify the hand into 12 blocks based on the relationship of current pixel with its surrounding pixels in the depth image. Finally, the position of the joints between the neighboring blocks are used as the pattern to search in the gesture library and then the gesture is recognized.