Cross-Scene Crowd Counting via FCN and Gaussian Model
Hao Liu, Yadong Li, Zhong Zhou, Wei Wu · 2016
Cross-scene crowd counting is a challenging task where no laborious data annotation is required for counting people in new target surveillance crowd scenes unseen in the training set. The performance of most existing crowd counting methods drops significantly when they are applied to an unseen scene. To address this problem, we propose a Full Convolutional Neutral Network(FCN) for person detection, which has a reliable performance for cross-scene application. The network evaluates every pixel for the possibility of being part of body and further derives a confidence coefficient map. Then we establish a weighted adaptive human Gaussian model according to the different sensitivities for different human part in the map. Lastly we proposed an algorithm mapping the confidence coefficient map to the crowd counting and density distribution with the Gaussian model. Our method doesn't need lots of annotations for new scene. Extensive experiments on the proposed datasets demonstrate the effectiveness and reliability of our approach.