The visual detection features of cnc processed images include color features, texture features, spatial relationship features, and shape features. Color features are based on all features belonging to image pixels. Texture features are statistically calculated in an area containing multiple pixels. Spatial relationship refers to the spatial position or relative orientation relationship between multiple objects in an image. Shape feature refers to a specific shape composed of a set of geometric elements (points, lines, surfaces) with certain topology relationship on the workpiece. As important visual information of objects, shape features are stable attribute representations of objects.
According to the comparison between different visual features and the analysis of the dynamic benchmark above, the shape feature conforms to the characteristics of the dynamic benchmark, so the shape feature is selected as its visual feature. Generally, shape features can be represented as shape-based contour features and shape-based region features, including corners, edges, straight lines, curves, and regions. If there is interference in the edge, a large number of false edge points will be generated, which will affect the edge feature extraction. At this time, it is very important to find a suitable image processing algorithm.
Correspondence Analysis and Feature Extraction Method In the process of CNC machining, when the machining reference is the corner of the workpiece, it appears in the form of corner points in the visual image. Among the image corner visual detection algorithms, the template-based method is the most widely used, including Harris algorithm, Susan algorithm, FAST algorithm and SURF algorithm.
Among them, the Susan algorithm has good robustness to noise, and also has the invariance of light intensity change and rotation invariance. It uses very few parameters, so it requires less computation and storage. Therefore, the Susan algorithm is used to extract the corner coordinates, that is, the position of the dynamic reference.
When the machining reference is the axis or centerline of the revolving surface, it appears as the center of the circle in the visual image, but the center of the circle is not real, so the extraction of its visual features involves the extraction of the edge of the circle and curve fitting. Obtaining the center coordinates is the machining dynamic reference position. Among the commonly used edge extraction operators, Canny operator has higher positioning accuracy for single-pixel edge than other edge detection operators, and has better anti-noise ability.
