Corner Detection Based on Estimated Eigenvalues

corner detection, autocorrelation matrix, Harris Plessey operator, Förstner operator, Kitchen and Rosenfeld operator, eigen-values, image processing

The main goal of this paper is to introduce existing methods for corner detection which are suitable for HW implementation. In paper is introduced a new method for corner detection based on estimated eigen-values of auto-correlation matrix and comparison of mentioned detectors.

Corner detection is important part of many image analysis methods and the result of corner detection is essential for performance of stereo vision, augmented reality, etc. This paper proposes a new method of corner detection based on autocorrelation matrix estimated eigen-values and compares the new method with several frequently used methods The new method is superior to the more traditional ones as it is one-pass, robust, and also because it is easy to implement in programmable hardware. The main novel part of the method is that for determination of corner maxima it directly estimates eigen-values of autocorrelation matrix. Exact eigen-values computation is time consuming and it is obtained by solving the second order polynomial. However, estimation of eigen-values can be obtained through solving first order polynomial which gives good results and is faster than the known methods.