摘要：AbstractThe driver of high-speed trains usually is required to perform certain gestures to confirm the signals before implementing some operations, which is an essential validation for driving safety. However, the accuracy of gesture recognition is difficult to guarantee due to the jamming background and limited perspective. In this paper, the features of the side view and vertical view are integrated to assist classification decisions. Firstly, point clouds of the gesture are generated with RGB-D data and then projected onto two orthogonal planes to reconstruct the side and vertical view of the gesture. Secondly, multiple-view 3D Convolution Neural Network architecture is proposed with three branches of Convolution Neural Network. Combined with the front view obtained by frame difference, the model learns convolution features from three aspects of the gesture. Further, multiple-view classification results are adaptively fused to acquire the final decision. Experiments show that our approach is superior to the state-of-the-art gesture recognition methods on challenging dataset.
关键词：KeywordsDriver Gesture CongnitionTrain Operation SafetyHuman-centred ComputingModeling of Human PerformanceCognitive Systems Engineering
