| Summary: | 碩士 === 國立東華大學 === 電機工程學系 === 106 === The accurate sound source localization is to find the intersection of estimation range by TDOAs. However, the sample rate is not infinite in reality, the sample error in TDOAs makes the estimation range shift. The above error causes the source positioning offset or even no solution in geometric positioning. In order to solve the problem aforementioned, a novel geometric approach based on multiplanes is proposed.
In this thesis, the difference which the estimated range change from the conventional hyperboloids method to the multiplanes is proposed. By using the equation of the sensor array vector and its TDOAs. Geometrically, multiplanes are more likely to have intersections than hyperboloids, and the effect of error caused by sample error is lower, and the estimation of sound source position is less to have an empty set solution. And a discriminant is added in this study in the case of more than five sensors, t can be used to select a suitable TDOAs to estimate better solution.
The experimental simulation proves that this algorithm can effectively improve the estimation success rate and stability of geometric positioning, and reduce the situation that the original geometric methods cannot be performed under the influence of sample error, and in the case of multiple sensors. The estimation accuracy will also be improved.
|
|---|
