This forum is disabled, please visit https://forum.opencv.org
 | 1 | initial version |
I needed a robust ellipse fit out of contour points from OpenCV.
I found this paper:
https://www.princeton.edu/~kulkarni/Papers/Journals/j074_2010_YuZhengKulPoor_EURASIP.pdf
It uses a combined outlier detection. At first proximity is checked. In the second step a model based detection is used. It seems to be much faster and more robust than RAMSAC.
Because the findContour algorith from OpenCV gives me only points with direct proximity, I only had to programm the model based algoritm. For 5000 points the calculation time is about 1ms for choosing inliners, with SSE Intrinsics its about 0,2ms on an Intel Q9300 processor.
