The goal of this work was to develop a real-time Instrument-to-Midi-converter which transforms a monophonic music signal into a Midi stream. The converter measures the pitch of the music signal and converts it into Midi NoteOn and NoteOff commands. A Windows-PC with a Trimedia DSP Board in a PCI slot has been used as a platform for this converter. The DSP does the real time processing work, while the PC is used to control the DSP and to write the computed notes to the Midi-interface of its soundcard.

For pure monophonic signals, the converter works pretty good. This is especially true for all the tested windinstruments. When there some reverb is present in the signal, the detection sometimes may fail. The same problem may arise using a sustaining guitarstring.

The detection of the pitch in the input signal is done using a zerocross calculation. The converter determines the amount of samples between two zero crossing. This number of samples can be converted to the fundamental frequency. A filterbank with nine bands is used to eliminate the disturbing harmonics during the calculation. Two different filter types can be choosen: lowpass and bandpass. In order to eliminate false detections, the converter can be individualy configured for any instrument by three parameters: Attack, Release and Threshold. These parameters can be controlled in real-time by the PC's GUI.

The aim of this diploma project is, via digital image processing, to detect a defined person wearing a coloured T-shirt. If the person who, for example, may be standing on a theatre stage is detected, his position can then be calculated. With the two dimensional position established, the volume parameters of the surround-system can be controlled. The members of the audience who are within equi-distance of the loudspeakers have the illusion that the sound source is synchronised with the movements of the actor. Up to now, this problem could only be solved by using a joystick to follow the movements of the person.

We elected to use a VASP System (Video Controlled Audio Surround) to complete our task. This system is based on a previous diploma project done in 1998. We have extended this project by the T-shirt recognition algorithm. First, we simulated and developed the image processing algorithms using real pictures. For this purpose we used the Matlab image processing Toolbox. We developed solutions for the recognition of colours, detection of movement and measurement of colour surface dimensions. We were not able to simulate the dynamis charactersitics of the recognition in Matlab. Therefore, we completed this after the implementation of the image processing algorithms in the real time system.

The test phase showed that the T-shirt recognition algorithm only has the predicted restrictions, which are based on system limitations. Now, an individual can be detected and his position calculated. The recognition algorithm also functions with several individuals, and even ones wearing similar T-shirts.