Pedestrain Tracking And Counting In Survilliance Videos

Pedestrain tracking and counting in video are significant research topics in computer vision.Recently,Correlation Filter(CF)based methods have demonstrated excellent performance for visual object tracking.However,CF-based models often face one model degradation problem: With low learning rate,the tracking model cannot be updated as fast as the largescale variation or deformation of fast-motion targets;As for high learning rate,the tracking model is not robust enough against disturbance,such as occlusion.To enable the tracking model adapt with such variation effectively,a progressive updating mechanism is necessary.In order to exploit spatial and temporal information in original data for tracking model adaptation,we employ an implicit interpolation model.With motionestimated interpolation using adjacent tracking frames,the obtained intermediate response map can fit the learning rate well,which will effectively alleviate the learning-related model degradation.The evaluations on the benchmark datasets KITTI and VOT2017 demonstrate that the proposed tracker outperforms the existing CF-based models,with advantages regarding the tracking accuracy.Due to a wide range of various application scenes,robust crowd counting is still quite difficult and the performance is far from being satisfied.In this paper,we propose a novel robust crowd counting method by introducing a weakly supervised crowd-wise attention network.The proposed work improves the counting accuracy and robustness by: i)Weakly-supervised crowd segmentation.With a generated segmentation label using motionguided region-growth,both the appearance feature of one labeled image and motion features abstracted from its adjacent unlabeled frames,are combined to implement weakly supervised crowd region segmentation,with which active crowd region can be finely perceived from different background disturbances.ii)More accurate spatial attention.We generate a spatial attention map based on the active crowd segmentation,which is used to reweigh the appearance feature to achieve attention-based density estimation.iii)A new temporal-densely-annotated crowd dataset.We introduce a dense temporal annotated video crowd dataset in order to complement lack of crowd datasets with rich motion information.Evaluation on the widely used World Expo'2010 and our proposed surveillance crowd video dataset(SCVD),shows that the proposed work can achieve state-of-the-art performance on both the accuracy and robustness.