| Title | Position Informed Convolution for Multi-Agent Line Detection |
| Author | Petr Babkin |
| Consultant | Nikita Shubin, PhD |
| Year | 2024-2025 |
Research project by student of MIPT at Intelligent systems department.
This paper introduces a novel approach for robust curve detection in images, a task often hindered by the high degree of parametric freedom in curve representation. Our method utilizes a multi-agent system, wherein individual agents employ the Hough transform to identify potential curve segments. To effectively consolidate these partial detections and account for their respective spatial contexts, we propose the Positional Informed Convolution (PIC) layer. This novel layer extends traditional convolutional operations by explicitly encoding the spatial location of input feature maps, thus enabling a more sophisticated and contextually aware aggregation of agents’ outputs. The effectiveness of our proposed approach is validated through experiments on a custom-built synthetic dataset, where we demonstrate significant improvements in curve detection accuracy and robustness compared to conventional methods.