Bidirectional Convolutional Long Short-Term Memory-based Human Action Recognition on an Airport Apron

Abstract:
With the rapid development of the air transportation industry, the security and management problems of large-scale hub airports have become increasingly prominent. In order to achieve effective monitoring on the apron under the condition of low visibility and prevent unsafe accidents, this paper adopts the strategy of tracking and preprocessing before recognition for the non-cooperative individual targets in the apron. Considering the characteristics of infrared monitoring system, the proposed method can maximize the use of effective features while reducing the cost of calculation, and can be applied to the actual situation of multiple targets and complex background. Furthermore, the bidirectional recurrent neural networks-based model is built to handle recognition tasks. The recognition accuracy is 92.2% under the infrared dataset of airport apron. This model performs well in behavior recognition tasks with challenging scenes.