Detection of Object Carried Using Spatio-temporal Pattern and Local Directional Pattern Descriptor

A novel method of carried object detection based on silhouette width and spatio-temporal analysis is proposed in this paper. The width vector of silhouette is calculated to characterize the spatial feature and the width image is adopted to preserve the temporal features. The width image which is considered as the texture image represents the walking sequence of pedestrian in a grey-level image. For analyzing the texture features, the local directional pattern code is used to encode the local properties and the local texture pattern descriptor is applied to capture the global characters. Chi-square distance equation is exploited to measure the difference between descriptors. Experimental results show that the width image is an effective and efficient representation, the local directional pattern descriptor is robust and insensitive to noise, and our method is effective.