Human typical action recognition using gray scale image of silhouette sequence

Correct recognition of human action is the basis for home service robots. Human action recognition is a research hotspot unsolved completely at present. This paper presents a new way of human action recognition. RBF (Radial Basis Function) neural networks-based methodology using silhouette as a representative descriptor of human posture to achieve daily human actions recognition is proposed. The silhouette features of an action sequence are transformed to a grayscale image after building an adaptive background model. Then the gray scale image dimension is reduced by the DCTs (Discrete Cosine Transforms) and the radial basis function neural network is employed to recognize human actions. Finally, the experimental results are provided to show the effectiveness of the proposed method.

Figure optionsDownload as PowerPoint slideHighlights
► Use the contour of human posture to recognize human actions.
► Achieve action recognition by means of the vertical and horizontal projection of the silhouette sequence.
► Transform time-sequential images to a grayscale image feature vector.
► Apply DCT to reduce the gray scale image dimension and radial basis function neural network to recognize human actions.