Service Action Recognition in Power Supply Business Hall with 3D-Fused ConvNet

For the purpose of improving service quality, video surveillance systems are widely used to standardize process in power supply business halls. If employers check ensure predefined staff behaviours, it will be characterized as time-consuming. In recent years, great progress has been made intelligent action recognition using Convolution Neural Networks (CNNs). However, due small range staffs' motion and similar scene information halls, performance traditional CNNs recognize actions, e.g. bowing, standing sitting, is general. rate, this paper proposes a 3D-fused Convolutional Network (ConvNet) for actions recognition, which focuses on detecting typical one person customer with well-segmented clip. The clips sent input ConvNet recognition. consists two base learners, optical flow learner RGB learner. Both learners use 3D (C3D) architecture. Specifically, can capture features while viewed key part eliminate influence background, especially scene. Furthermore, prediction scores weighted by softmax function according each Finally, fused obtain result, namely specific staffs videos. experiment result shows that proposed method achieves 92.41% accuracy dataset hall.