Date of Publication :27th December 2017
Abstract: Recognizing human action from video still remains a problem. Robust solutions to this problem have applications in various domains such as surveillance systems, human-computer interaction, smart home health-care systems and control free gaming systems etc. In this paper, an approach for human action recognition based on neural network is presented. In this neural network approach, motion vector feature is used for action recognition. The features are extracted directly from the video without any pre-processing. Since preprocessing is not necessary for the proposed approach computation time has been minimized as compared to that of the existing system. Then the network is trained using scaled conjugate gradient backpropagation method. The neural network is trained with the motion vector features. The training is based on the number of hidden neurons, the percentage of sample data taken for training, validation and testing. This system provides an efficient result with a minimum number of hidden neurons and training data. The proposed approach is tested on Weizmann dataset that consists of 10 actions providing 9 videos per activity.
Reference :
-
- Lacher R. C, Hruska S. I, and Kuncicky D. C, “Backpropagation Learning in Expert Networks”, IEEE Transactions on Neural Networks 3, pp. 62–72, 1992.
- Battiti R, “First- and second-order methods for learning; between steepest descent and Newton‟s method”, Neural Computation, Vol. 4, pp. 141–166, 1992
- Baldi P. F and Hornik K, “Learning in linear neural networks: A survey”, IEEE Transactions on Neural Networks, Vol. 6, pp. 837–858, 1995.
- Weber M, Welling M, and Perona P, “Unsupervised learning of models for recognition”, Lecture Notes in Computer Science, Jan 2000.
- D.Pham, S. Sagiroglu, “Training multilayered perceptrons for pattern recognition: a comparative study of