Least squares multi-level approach for centroid detection in active triangulation

Active triangulation has been successfully applied in numerous applications such as distance measurement, profile inspections and pose estimation in robotics. Its accuracy depends on many factors with locating laser centroid being one of them. The error generated in locating the laser centroid within the image is propagated to the system output with consequences in decreasing performance. In this paper, we propose an algorithm for reducing error and hence improving the measurement accuracy. The approach applies gamma square law as a preprocessing techniques to reduce effects of background noise and speckles from the laser. Pixels clusters are extracted from the image by searching for the pixels with intensity levels within a pre-defined value and then, the Region Of Interest (ROI) is located by looking for an object shape similar to an ellipse within the extracted clusters. Each row and column of the ROI is divided into three sub-pixels classes. Least-square method is used to fit linear regression into each subclass. The standard deviation computation from this method is used to locate the centroid of the laser spot with better accuracy. The experimental results obtained using this method show consistent and improved results with least error deviation.