Smart Feature Fusion and Model for Human Detection

Extraction of discriminate and accurate features is challenging to precise the statistical data on monitoring people. It still remains an active research due to various variations as inter class and intra class, lighting challenge, static and dynamic occlusion. To tackle this variation and occlusion issue, this paper proposes to combine the differential gradient and statistical Tamura features with joint histogram. In addition, these extracted smart features use actually to detect people by using the gradient feature descriptor and a statistical feature detector. The model fusion of human detection creates by combining two models result (Grammar model and Poselet model) with the adaptive threshold weighted non-maximum suppression algorithm. The system presents a powerful fusion insight to capture the stronger occlusion parts and several variations of the foreground people. To compare the performance with the state of the arts, the public Pascal VOC 2007 Dataset is used. The outperformed result of this work proofs our concern.