Tracking Key Technology Research Of Video Audio Fusion Based On The Head Of Robot

Sound and image information as human perception is not only the main way to theenvironment,but alsothekeytechnologyofvideoaudionavigationbased onhumanoidrobot. Inorder to improve the navigation capacityof robot, it is an important part of research using audiovideo technology. The development of robot video and audio will not only greatly promote thedevelopment of intelligent system, also broaden the scope of the study and application areas ofintelligentmachines.Thetarget trackingis afoundational needofhumanrobot,so themainworkofthis paperisnamely target tracking based on audio video fusion. Video tracking is more mature than audiotrackingwhichisstillintheresearchstage,sothataudioisemphasesofthispaper.Thereforethetestflatismainlybasedonaudio.At first, in order to solve this noisy environmental background for the limitations of thispaper, this paper put forward a new approach of de-noising that is the second-decompositionwavelet global threshold. Wavelet transform is the basic theory of this approach, and thethreshold method of de-noisingis used. This approach differs from conventional methods in thatitdecomposeshigh-frequencyparttwice.Testresultsshowthatthisalgorithmimprovedsignaltonoise ratio, and wiped off the most noise with the intact effective energy. Therefore theinterference problem of voice signal is absolutely solved. At the same time, this part built up agoodfoundationforpitchdetecting.There are many methods that tracks pitch detecting of speech signal, however, thesemethods proposedhave not onlyadvantages but also disadvantages. Soa newalgorithm is givenin this paper, that is the fusion auto-correlation function and average magnitude differencefunction. In general, this method is applicable to the noisy environment of real-time processing.Thisstephasplayedakeyroleinspeechseparationandmulti-targettracking,testresultsshowedthat this method behaves robustly very much, and simultaneously pitch detecting is very rapid.We also can get a conclusion that peaks are more obvious and the judge of surd and sonant ismoreaccurate.Finally, ImportanceParticleFilteris usedas target trackingalgorithm inaudiovideofusion.In this algorithm, audio and visual information are in a symmetric manner so that they cancompensate for each other to a better extent. This algorithm is more robust than trackingalgorithm based only on visual and audio information, and it is also robust to light change,backgroundchange,occlusiontosomeextent.