Estimating count and density maps from crowd images has a wide range ofapplications such as video surveillance, traffic monitoring, public safety andurban planning. In addition, techniques developed for crowd counting can beapplied to related tasks in other fields of study such as cell microscopy,vehicle counting and environmental survey. The task of crowd counting anddensity map estimation is riddled with many challenges such as occlusions,non-uniform density, intra-scene and inter-scene variations in scale andperspective. Nevertheless, over the last few years, crowd count analysis hasevolved from earlier methods that are often limited to small variations incrowd density and scales to the current state-of-the-art methods that havedeveloped the ability to perform successfully on a wide range of scenarios. Thesuccess of crowd counting methods in the recent years can be largely attributedto deep learning and publications of challenging datasets. In this paper, weprovide a comprehensive survey of recent Convolutional Neural Network (CNN)based approaches that have demonstrated significant improvements over earliermethods that rely largely on hand-crafted representations. First, we brieflyreview the pioneering methods that use hand-crafted representations and then wedelve in detail into the deep learning-based approaches and recently publisheddatasets. Furthermore, we discuss the merits and drawbacks of existingCNN-based approaches and identify promising avenues of research in this rapidlyevolving field.
展开▼
