Sub-pixel Edge Detection Based On Hyperbolic Tangent

With the innovation of the science and technology, the application of visual measurement in various fields being widespread attention and concern gradually. Therefore, taking into account the detection accuracy, speed, stability and other kinds of factors, the visual measurement technique which has many advantages, has been replacing the traditional detection methods step by step, and turns into a widely used means of measurement.As the basis for visual measurement, edge detection and corner detection affects the accuracy of the measurement directly. Focusing on improving detection accuracy, while improving the speed and stability, here present a visual measurement method based on hyperbolic tangent in this paper.First, in order to clear the factors which impact on the positioning accuracy of the edge, the edge technique has been in-depth researched in this paper, such as the methods of moment, interpolation and fitting. According to a description of the actual intensity distribution of the edge could be approximated well by a hyperbolic tangent function, and combine with fit principle, the hyperbolic tangent function was selected to implement the proposed edge detection algorithm. This method is tested with real data:1. The feasibility of the edge detection algorithm has been analysis through experiments, while the experimental data prove that this method both have speed and stability; 2. By changing the detection window size and parameters of fuzzy factor, analyze the factors that influence the method; 3. In order to compare the accuracy of this algorithm with Gaussian method, a direct evaluation of edge detection accuracy that based on fitting error has been proposed in this paper, which use the high-precision translation table carry the object move on the fixed direction under a certain step; The experimental data demonstrate that the method could achieve higher precision and faster speed. Then, for the detection of X-corner calibration plate, not only improved Harris and Susan method, but also Forstner method have been introduced in detail. According to the characteristics of the corners (that is made from the interaction of two or more straight lines) and the edge detection algorithm which based on hyperbolic tangent, a sub-pixel corner detection method would be proposed. In order to test the new algorithm, measure the pixel distance and actual distance of any gird which belong to the calibration plate. Compare with the improved Harris method, experimental data show that the algorithm which proposes in this paper could achieve very good precision, and can effectively suppress noise.Finally, in order to verify the accuracy of this method, apply the sub-pixel detection to detect different sizes of stander gauge blocks. Using the detection results to calibrate the CCD camera's parameters, and then combine with the edge detection to find the size of the objects. According to the comparison of the experimental data and stander size could be found that, the method can extract the stander gauge blocks accurately.Therefore, it could reach a very high speed, stability and accuracy detection when use this algorithm in visual measurement, which could provide a viable reference to the on-line non-contact measurement.