Vision based Human Activity Recognition using Deep Transfer Learning and Support Vector Machine

In this computer era, the interaction between human and machine is an extremely interesting area due to the advent of deep transfer learning and the availability of vision based human activity recognition datasets. Transfer learning is one of the decisive innovations in the field of deep learning due to its promising results. Therefore, this paper proposes the deep transfer learning method to extract the features in conjunction with a multi-class support vector machine for the classification of human activities using RGB videos. A fine-tuned pre-trained convolutional neural network based on Visual Geometry Group network 19 (VGGNet-19) has been applied to extract the visual features from the RGB videos. Next, a multi-class support vector machine is used to classify the extracted features into the different categories of human activities. The proposed approach is evaluated on a popular sports activity dataset such as: UCF Sports Action dataset. During experimentation the proposed approach achieves 97.13% recognition accuracy on this dataset. The experimental results also show the effectiveness of the proposed approach over handcrafted feature engineering methods such as histogram of oriented gradient and optical flow.