Abstract
In recent years, video image recognition technology has become widely used in various fields. With the recent release of hybrid sensor devices combining RGB camera and depth sensors, the accuracy of motion analysis has greatly improved. Currently, research on motion analysis has been centered on human motion analysis and gesture detection. On the other hand, research areas such as ethology, or study of animal behavior, has large potential for automatic image analysis and detection using video cameras, but there has been few research on applying hybrid sensors for improved animal recognition. This is partially due to the design of the current hybrid sensor software which is targeted primarily for human gesture detection, and requires costly modeling of targeted animals to achieve accurate recognition using the same approach. For this research, we propose a meta-heuristic learning approach for animal recognition using a hybrid sensor camera. We train an artificial neural network for a particular animal recognition using the depth sensor array input for the network input, and verify the validity of the proposed method for dog recognition.
