Enhanced automated body feature extraction from a 2D image using anthropomorphic measures for silhouette analysis

Abstract

Approach that derives human metrology measurements from 2D images of silhouettes.Dynamic background segmentation to support silhouette detection with realistic backgrounds.Parallel feature extraction method based on an anthropometric search for online processing.Validity of the improved silhouette detection approach for silhouette identification. Using mobile systems (e.g., smart phones, tablets, mobile robots), human metrology (HM) is a soft biometric that can be beneficial for human detection and identification from a 2D image, as long as it can be done online and in open conditions. This paper presents an approach that derives HM measurements from 2D images of silhouettes. The approach enhances an algorithm for automated body feature extraction from a 2D image in front of a black background, by using anthropomorphic information to extract, online and in parallel, 20 front and 13 side measures out of 45 front and 24 side features, increasing the number of features and improving processing time. Recognition and identification results are presented with both uniform (black) and real backgrounds, for distances ranging from 1m to 6m.