Moving cast shadow detection using scale-relation multi-layer pooling features

Abstract

Moving cast shadows detection and removal are indispensable for object detection and are the problems in visual surveillance applications which have been studied over the years. However, finding an efficient model that can handle the issue of moving cast shadow in various situations is still challenging. Unlike prior methods, we use a data-driven method without strong parametric assumptions or complex models to address the problem of moving cast shadow. In this paper, we propose a novel feature-extracting framework called Scale-Relation MultiLayer Pooling Feature Extracting (SMPF) which includes two main tasks: (1) Scale-Relation Scheme (SRS), (2) Multi-Layer Pooling Scheme (MLPS). By leveraging the scale space, SRS firstly decomposes feature images of each shadow properties into various scales and further considers the relationship between adjacent scaled feature images of each shadow properties to extract the scale-relation features. Then, we design the multi-layer pooling scheme (MLPS) to integrate the features in a local region and to reduce the dimension of extracting features. After that, the density map is generated for various properties of shadow with low dimension. Finally, to seek the criteria for discriminating moving cast shadow, we use random forest algorithm as the ensemble decision scheme. The main contributions of this study are (1) we design the features with multi-scale which can provide abundant information to describe the moving cast shadow, (2) the multi-layer pooling scheme generates the density map to integrate and reduce the dimensions of features. Experiments on the popular benchmarks and the proposed dataset with benchmarks demonstrate that the proposed method can achieve the performances of the popular methodologies.