Deepsim AI Porject for Human Behavior detection

Human behavior recognition plays an important role in social interaction in different fields

This is an international cooperation research project between Deepsim Intelligence Technology Inc. (Deepsim) and Yantai University. This project was initiated by Dr. Shouke Wei in Deepsim in September 2020, and it was completed in June 2022. Human behavior detection plays an important role in social interaction in various fields. This project proposed a skeleton key points and two-layer bidirectional LSTM-based Seq2Seq model (SB2_Seq2Seq) for human behavior recognition. An experiment was conducted to compare the proposed SB2_Seq2Seq with other five skeleton based models, including three LSTM-based Seq2Seq models with two-layer unidirectional LSTM (SU2_seq2seq), single-layer bidirectional LSTM (SB1_seq2seq) and single-layer unidirectional LSTM (SU1_seq2seq), respectively, and the traditional LSTM and CNN models. In addition, SB2_Seq2Seq were also compared with RGB video frame-based B2_Seq2Seq model (RB2_Seq2Seq). The UCF50 dataset was used for model evaluation, where 60%, 20% and 20% for model training, validation and testing. The experimental results show that the SB2_Seq2seq model achieves 93.54% accuracy with 0.0214 Mean Squere Error (MSE), suggesting that SB2_Seq2seq has better recognition performance in terms of accuracy and MSE than all the other six models. Besides, compared with RB2_Seq2Seq, SB2_Seq2seq also greatly increases the training speed.