Infrared video behavior recognition algorithm based on AGX Xavier environment

Analyzing human behavior recognition for infrared video is important since crimes are more likely to occur in nighttime environments. Therefore, this paper proposes an infrared video behavior recognition algorithm based on AGX Xavier, an edge device, using a TSM module embedded in a 2D neural network with temporal modeling capability, which can significantly reduce the number of model parameters and computational effort to achieve deployment on a lightweight edge device with cameras. Online behavior recognition is achieved. The method's accuracy is 91% on the dataset captured by the subject's active infrared camera. It is tested in real application scenarios, and the online recognition mode has low latency and high speed to meet the real-time behavior recognition requirements.