Gesture recognition based on spatiotemporal histogram of oriented gradient variation
Seiji Kojima, Wataru Ohyama, Tetsushi Wakabayashi · 2017
A fine-grained gesture recognition method based on spatiotemporal representation for cooking activities is proposed. Cooking is one of common housework activity in daily life. Supporting cooking using video-based gesture recognition can contribute to improve our quality of life. A cooking gesture recognition method which employs a spatiotemporal representation for both appearance of a cooker and surrounding kitchen utensils. Our proposed method employs Spatio-Temporal extension of Histogram of Oriented Gradient Variation (ST-HOGV) which can represent not only appearance and temporal change of independent objects but locations of these objects. Performance evaluation experiment using ACE dataset shows that recognition accuracy of 76.4% is obtained and the KSCGR evaluation score achieves 73.5%. While the proposed method does not require any a priori knowledge, the performance is comparative other gesture recognition method with a priori knowledge.