Research on Ship’s Officer Behavior Identification Based on Mask R-CNN

The negligence and inattention of the navigators on duty are the main causes of ship accidents. Based on the CCTV records of the bridge, the identification on human behaviors on duty has high value and significance. In this paper, a human behavior recognition method based on Mask R-CNN and a knowledge-based model is proposed. At the very beginning, the images of bridges collected by the camera are used as the input, the keypoints are extracted through a convolution neural network. Subsequently, the feature distribution of the keypoints of the k-nearest neighbor algorithm is then analyzed by clustering, which realizes the real-time online recognition of common behaviors such as walking, standing, sitting, steering and looking through the mobile phone, looking out and napping. At the final step, the proposed method was validated on the samples of the ferry ship from Zhoushan, Zhejiang, China. In the validation, the one-time recognition accuracy of this method was 82%, and the rolling recognition rate had reached almost 99%, which might significantly improve the safety of navigation.