ACTION RECOGNITION FOR GROOMING BEHAVIOR IN RATS USING SUBSAMPLED SEGMENT RECURRENT

• Main Authors: Wei-Wei Gao, 高唯唯
• Other Authors: Chien-Cheng Lee
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/vp4344

碩士 === 元智大學 === 電機工程學系乙組 === 107 === Grooming behaviors of rats can be used to reflect its states of physiology and psychology. Here, we develop a grooming detection method using deep learning algorithms with image processing. We propose a novel approach that separates the recognition procedure into two phases. In the first phase, top-view video images over rat’s cage are obtained. Frame-to-frame image differences are then calculated to form a time series that is filtered into absolute non-grooming and candidate grooming clips. In the second phase, our proposed model subsampled segment recurrent ConvNets classify the movement patterns on the candidate grooming clips. Results show that rat grooming behavior can be detected at high accuracy over 98%.