eduzhai > Applied Sciences > Engineering >

Complex Human Action Recognition in Live Videos Using Hybrid FR-DL Method

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: Automated human action recognition is one of the most attractive andpractical research fields in computer vision, in spite of its highcomputational costs. In such systems, the human action labelling is based onthe appearance and patterns of the motions in the video sequences; however, theconventional methodologies and classic neural networks cannot use temporalinformation for action recognition prediction in the upcoming frames in a videosequence. On the other hand, the computational cost of the preprocessing stageis high. In this paper, we address challenges of the preprocessing phase, by anautomated selection of representative frames among the input sequences.Furthermore, we extract the key features of the representative frame ratherthan the entire features. We propose a hybrid technique using backgroundsubtraction and HOG, followed by application of a deep neural network andskeletal modelling method. The combination of a CNN and the LSTM recursivenetwork is considered for feature selection and maintaining the previousinformation, and finally, a Softmax-KNN classifier is used for labelling humanactivities. We name our model as Feature Reduction & Deep Learning based actionrecognition method, or FR-DL in short. To evaluate the proposed method, we usethe UCF dataset for the benchmarking which is widely-used among researchers inaction recognition research. The dataset includes 101 complicated activities inthe wild. Experimental results show a significant improvement in terms ofaccuracy and speed in comparison with six state-of-the-art articles.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×