Research on LED die geometric parameter measurement based on shape recognition and sub-pixel detection

Existing graphic element detecting algorithms possess several drawbacks such as high complexity, low measuring accuracy and poor robustness to image degradation. A two-stage hybrid approach is proposed to solve these problems in detecting and measuring LED die geometric parameters. In the first stage, the LED die binary image is obtained by adaptive multi-threshold segmentation, and the LED die graphic elements are transferred into connected components, which are composed of some geometric primitives such as point, line, circle, arc, etc. On the basis of analysis of the connected components in LED die binary image, the shape feature parameters and the similarity measures of graphic elements are calculated to identify their affiliation and spatial topology, and the bounding boxes of target graphic elements is extracted. In the second stage, an improved random Hough Transform based on least square fitting is used to measuring LED die geometric parameter: firstly, the ranges of HT parameters for geometric primitives are assured by histogram scanning the edge points within the bounding box; secondly, by the aid of direct least square fitting to target geometric primitives, an iteration procedure of collecting the probable edge points and refitting the primitive is introduced to extract the final primitive parameters; at last, the shape and location of LED die graphic element are obtained by geometric operations. This hybrid approach is applied in LED die test experimental equipment for motion calibration and LED die geometric parameters calculation, and the result of experiments demonstrates that the proposed approach possesses high precision and robustness.