Video Face Detection And Target Tracking Based On Deep Learning

With the rapid development of science and technology,computer technology plays a very important role in today's social life.The ability of computers to process visual information largely compensates for the shortcomings of human beings,thus making computer vision research One of the most attractive research directions today,computer vision analysis is a highly intersecting discipline that includes graphic image processing,pattern recognition,and artificial neural networks.The purpose of computer vision research is to make the computer perceive the geometric information of the object in the environment through its analysis of the video image,including its shape,position,posture,motion,etc.Nowadays video image analysis technology in image segmentation,behavior recognition,target Achievements in tracking and other fields are increasingly being used in many areas such as corporate security,aerospace,smart transportation,and home security.This paper mainly studies the face detection and recognition tasks under non-restrictive conditions,and uses target tracking to assist the face.Through the research and summary of the related technical knowledge of face detection,target tracking and convolutional neural network,a large number of volume-based volumes are completed.The face detection and recognition and target tracking experiments of the neural network are applied to the enterprise products and being continuously optimized.The areas of research focused on this paper include: 1)Designing a fast and accurate detection network.After a lot of experimental analysis,this paper designed a fast deep convolutional neural network to speed up the detection speed while achieving a balance between speed and accuracy.At the same time,this paper uses the dataset of WIDER FACE,which is the joint development of the University of Hong Kong,Shang Tang Technology,Shenzhen Institute of Advanced Science,Chinese Academy of Sciences and the method is tested on the face dataset FDDB to verify the feasibility of the training model.2)A recursive neural network-based target tracking algorithm is proposed,based on the long-short-time memory network combined with spatiotemporal context information.Link the information of the continuous video sequence,and learn the robustness of the initialization parameter enhancement model for future frames,Testing trace model performance on VOT datasets;3)Design the security system based on face recognition and target tracking in complex scenarios: This paper designs a multifunctional security system with face detection,face recognition,target tracking,stored data,real-time alarm and regional trajectory display through face detection and recognition.After many tests,it was displayed at the ISC(International Security Conference)and the National Cyber Security Week,and achieved good results.