Abstract:
This research purposes a digital image processing approach for gesture recognition, from single view video sequence at static and clutter background. Specifically, repeating path trajectories kicking and punching. At first, found the roughs motion area by project sum of difference frame profile both of vertical and horizontal. Next, we used template matching for extract especially limb motion and then representing the ROIs movement by MHI. The MHI can be encoded movement sequence into single frame but limited of MHI is cannot present movement that have repeating path trajectory type. We solved by searching the mid-action frame for separate video sequence into parts of video sequence and recognize kicking and punching by SVM. Our experiments are used other gestures for test recognition approach, there are foot thrusting and pushing. We tested by classify the target movement from 1 type of difference movement, and in addition to classify from 3 types of other gestures, after that ratio of input data for training, dividing movement to create parts of MHI and labeling history number are influence for the success rate, too. Resultant are showed 1 type of difference movement are highly rate that have highest recall rate of kicking and punching are 98% and 92%, respectively. The recall rate from 3 other gestures classification are 89% of kicking and 84% of punching.