Two Dimensional Human Pose Estimation for the Upper Body Using Histogram and Template Matching Techniques

The main purpose of human pose estimation is to estimate the size, position or orientation of the human body parts within the digital scene information. The estimation technique is directly influenced by the type of image feature to be used, its model representation and also the application of the system. This work focuses on estimating the size and position of the upper human body parts (head, torso and arms) in image sequences. Silhouette, edge and colour are selected as input features and are extracted using wavelet-based feature extraction. The pose of the head and the torso are estimated using histogram technique while the pose of the arms are estimated using template matching technique. The estimated pose of the upper body parts are then represented using rectangle shape. The performance of the pose estimation is measured in terms of correct size and correct position.