Research On Crowd And Vehicle Density Estimation Algorithm In Surveillance Video

In recent years,with the development of society and the improvement of people's economic level,various activities such as travel and large-scale gatherings are increasing,and people's demands for social security is also increasing.With the continuous growth of video data such as surveillance,digital research on crowd behavior and vehicle dredging is becoming more and more feasible.And studying the crowd and vehicles in the surveillance video is inseparable from the technical support of computer vision.The presentation and development of GPU(Graphics Processing Unit)provides convenience for computers to process large and complex image data.In addition to analyzing the meaning of population and vehicle behavior,this thesis also introduces the basic knowledge of convolutional neural networks,including the main components of the network,several classical convolutional neural networks and common deep learning fr-ameworks.At the same time,it introduces the current counting methods for population and vehicle,including some traditional methods based on physical analysis,and methods based on deep learning.The work of this thesis mainly includes three parts.First,a set of playground crowd data is given to enrich the image dataset in the field of computer vision.Second,different convolutional neural networks complete vehicle counting on the TRANCOS dataset and verify the generalization capabilities of these networks.Thirdly,aiming at the uneven distribution of crowd or vehicle density in the counting problem,an adaptive density learning framework is proposed.Adaptive density learning framework counts separately based on different scene densities of the image patch.The detection and regression counting methods are used for different image densities,and then the total target number is obtained by integration,which effectively improves the detection accuracy of the system.The proposed adaptive density learning framework can provide ideas and methods for the research of crowd and vehicle counting,and the proposed data set can be used as the data base for related research.