A Design Of Speaker Location And Tracking System Based On DSP

With the rapid development of science technology, the signals processing ofaudio and video based on DSP has become one of important subjects in technologyresearch. It becomes one point increasingly to audio and video technology for speakerdetection, locationand tracking. And it’s widely used in the field of video monitoring,video meeting and multi system. However, video, as single mode, could lead to severeerror of location and tracking due to weak light, even disable; the same thing wouldhappen to audio as a single mode, which could be a big challenge for audioperformance duo to high noise and severe reverberation. For solving this problem,audio and video fusion technology comes up. This paper studies mainly on targetspeaker’s location and tracking with the way of audio and video fusion technology,and then performance on the hardware platform.Firstly, we introduce the hardware platforms and choose one type for theoperation, then the DSP hardware platform with the chip DM6437based on DaVincitechnology is established as the support of audio and video information processingand fusion; secondly, we introduce deeply the mode and algorithm of time delay ofarrival(TDOA), and present a new way that three dimensional acoustic locationalgorithm based on coordinate conversion is used. It calculate the sound source’scoordinate in a new coordinate which is converted from original coordinate that thefour microphones are random. In the new coordinate, three microphones have theirown fixed position and the fourth one’s position is random. Then the sound source’scoordinate is calculated. Compared with routine TDOA algorithms, this methodincreases location precision effectively, and simplify the computational complexityafter coordinate conversion. And then Kalman filter and mean shift algorithm areintroduced in application of video. According to the experiment, their advantages anddisadvantages are proved. Therefore, the method of fusion of Kalman filter and meanshift algorithm is present, and is proved its robustness by experiment. Throughconfiguring the registers of CCDC, OSD and VENC in VPSS and setting TVP5146register, it works to collect and display video on hardware platform by chipDM6437’s processing. Finally, we fuse audio and video signals by Kalman filter, andburn the program to the hardware panel in order to make the panel locate and trackspeaker target independently. By combination both audio and video, it’s advantage isreduce the errors when single mode is used to locate and track, and make two modework together and get more precise and more stable performance of speaker targetlocation and tracking.