MDCN: Multi-scale Dilated Convolutional Enhanced Residual Network for Traffic Sign Detection.

Detecting small, multi-scale, and easily obscured traffic signs in real-world scenarios presents a persistent challenge. This paper proposes an approach that utilizes a multi-scale feature pyramid module to capture hierarchical features, facilitating robust detection of traffic signs across varying viewing angles and scales. To aggregate features at different scales and eliminate background interference, we employ a superposition of null convolution kernels with varying dilation rates, expanding the perceptual field from small to large. This effectively covers the object distribution across multiple scales while enhancing the resolution of the final output feature map for improved small target localization. Our method has demonstrated its effectiveness and superiority over several state-of-the-art approaches through extensive experiments conducted on two public traffic sign detection datasets.