Video Abnormal Action Recognition Based on Multimodal Heterogeneous Transfer Learning

Table 1

Test accuracy comparison of different methods.

Method	Testing accuracy (%)
Using the video frame images as input data	73.33
Using the human keypoints as input data	85.00
Using the normalized ratios of module length of SHAND as input data	76.67
Using the angles of SHAND as input data	85.00
Using the SHAND as input data	86.67
The multimodal heterogeneous transfer learning + SHAND	91.67