Fusion of Bag of Visual Words with Neural Network for Human Action Recognition

Recognition of human actions is one of the important tasks in various applications. It finds a number of applications in a range of areas. Some of the challenges confronted are noise that peeps in and shape of the action being performed by the subject. In this paper, a bag-of-visual-words is extracted with the help of Scale Invariant Feature Transform (SIFT) and optical flow. It represents spatial as well as time-dependent feature points. Sobel edge filter and median filtering is used to minimize shadow effect and suppress background noise respectively. The classifier used is multiclass Learning-vector Quantisation-based. The well-known KTH dataset is used as benchmark.