Identifying Intersections of Planes using Convolutional Neural Networks
Kirill D. Volkov, Pavel Alekseevich Panilov, T. Yu. Tsibizova · 2025
This research introduces a sophisticated methodology for analyzing the geometric characteristics of objects within two-dimensional images, emphasizing the detection of plane intersections. Unlike traditional algorithms, such as the Canny edge detector, which excel at generic edge detection but lack the specificity to discern intersection lines of planes, the proposed method leverages convolutional neural networks (CNNs) to enable advanced feature recognition. The hallmark of this approach lies in its ability to distinguish between positive and negative features of geometric elements, a capability critical for tasks requiring nuanced classification and contextual understanding. By integrating parallel data processing channels within the CNN architecture, the algorithm performs a multi-faceted analysis, isolating lines formed exclusively by plane intersections while filtering out irrelevant features, such as handwritten text or noise. This novel approach offers significant advantages in intelligent image classification tasks, particularly in environments characterized by high levels of visual complexity and noise.