The Study And Implementation Of The Algorithm For The Video-based Multiple Faces Tracking

. With the rapid development of network and video technology, the digital videomonitoring technology has been applied widely in many fields of the whole society,such as public security network system, road video monitoring system, as well asnumerous shopping malls, communities and building video monitoring systems. Thesesystems can record and store the video information of monitored scenes in real time,supporting the analyses of the following video content with large data volumes.However, monitoring and analyzing these massive video data needs lots of manpowerand material resources. Therefore, video automatic analysis has become the keytechnology of digital video surveillance. In this paper, the face tracking related to videoanalysis has been used as the research content and the key technology of trackingmethods has been researched and explored thoroughly to put forward an algorithm withbetter robustness for realizing the reliable face tracking.The specific research content and innovation as follow:(1) In term of feature selection, the wavelet feature for face tracking was adopted.The existing tracking algorithm is characterized by color. It is prone to jitter with poorstability and strong background noise. However, the wavelet feature is a kind of targetfeature with multi-resolution descriptive character. Its low grey level description of thewhole image combined with the high-resolution description of part detail regions candescribe targets better, reduce background interference, of low gray level of the wholeimage description, combined with high resolution on the details of the regiondescription can be used to describe the target and reduce the background interference.Combining with principal component analysis dimensionality reduction techniques, itcan also reduce the feature dimension thereby reducing the computation of waveletfeature matching process.(2) In term of face modeling, the modeling algorithm with memory character wasadopted. In the process of tracking the target face is not immutable. As a result, turningthe face will change the description result. Once transient block occurred the face willpartly or completely lost. The benchmark model algorithm has a memory of recent periods of time inside the target face records, which is presented by collective models ina weighted way. What’s more, the benchmark model assembling dynamic update canadapt to the change of target face smoothly, so it can track the target again when it isreappearing after a transient block.(3) This algorithm also combines the color feature to filter out invalid particles,which will exclude invalid particles by color features primarily in the establishment ofparticle samples and then calculate valid particles based on wavelet feature only. It canavoid wasting too much time on non-target areas and focusing on the process of the areathat is familiar with the target face.The intelligent monitoring system based on video scenes which was achieved bybasing on the face tracking algorithm mentioned above and using Microsoft VisualStudio2008integrated development environment combined with open source libraryCV2.2and interface library Microsoft Foundation Classes(MFC) can be used to trackmultiple target faces. This algorithm has a better robustness in terms of trackingstabilization and overcoming losing caused by sheltering.