HAND PART CLASSIFICATION METHOD USING DEPTH IMAGES AND APPARATUS THEREOF

摘要

The present invention relates to hand part classification method using depth images and apparatus thereof. According to the present invention, a hand part classification method using depth images may comprise: detecting hand part from cameras depth image; calculating feature value of each pixel included in the hand part, specifically calculating feature value for each pixel, by using the difference of average depth values for the two offset patches which are generated to have certain size, based on two points that are spaced apart by a set offset from the pixel within the depth image, as reference; classifying the hand part into multiple parts by inputting each of the pixels feature value into random decision forest, which is trained to output classification label corresponding to inputted feature value, in advance; setting central points for the multiple classified parts; and extracting a skeleton of the hand by using the set central points. According to the hand part classification method using depth images and apparatus thereof, a hand part can be effectively classified into multiple parts by using depth image only, thereby allowing accurate extraction of the hands skeleton information and improved recognition rate of hand movements.