Research On Human Motion Tracking System Based On Facial Recognition

The identification and tracking of a target persons has always been a hot research topic in the field of computer vision.With the development of artificial intelligence in recent years,deep neural networks have been involved in various fields.Some problems have better solution based on deep neural networks than previous solutions.This article mainly discusses the locking of the target character based on face recognition,and deals with the target character's motion tracking problem.This problem can be divided into two aspects.The first aspect is the identification of the target person.Since the overall characteristics of the target individual are not identifiable,the overall identification can easily lead to errors.Therefore,facial recognition is selected for the identification method of the target person.This article introduces the application of deep neural network in face recognition.Finally,it chooses facial recognition system of the SeetaFace which does not depend on any third party library functions and the code is based on C++.The open source protocol for this system is BSD-2.The second aspect is the issue of tracking moving figures.When it is recognized that the degree of similarity between the target face and the face within the sampling image reaches a certain level through the algorithm,it will automatically shift to track the whole torso of the moving figure.In the tracking part,several algorithms and the experimental data were compared.Aiming at the two problems of occlusion and similar color interference faced by traditional tracking algorithms,a Restricted CamShift Algorithm based on Directional Prediction(RCA-DR)is proposed.The classic CamShift as the optimization basis of RCA-DR,combines with the Kalman filter to prevent the target being occluded,and incorporates the inter-frame difference algorithm to prevent similar color interference between adjacent frames.Experiments have proved that the switching between face recognition and tracking algorithm is feasible.This solution prevents the similar color interference and occlusion of objects in the tracking process and guarantees the stability and robust of the tracking process.Accuracy and real-time performance have also achieved good results.