Framework for Traffic Sign Detection Based on Selective State Space Model

Traffic Sign Detection (TSD) is a crucial part for the autonomous driving system, guaranteeing safety and regulatory compliance. Despite the advancements in state-of-the-art detection models like You Only Look Once (YOLO), Convolutional Neural Networks (CNNs) etc., the need for enhanced detection of traffic signs under optimal conditions remains crucial. The study in this work introduces an enhanced methodology employing the Vision State-Space (VSS) block infused YOLO architecture to efficiently detect traffic signs with improved performance. The proposed framework integrates Symmetric Cross-Directional Selective Scan (SCDSS) and Double Convoluted Gated MLP (DCGMLP) with the VSS infused YOLO architecture, to enhance feature extraction and localization capabilities. The framework proposed is evaluated on the benchmark dataset of TT100K. The assessed metrics show that the proposed framework is capable of attaining better performance in TSD tasks as compared with many state-of-the-art methods. Overall, the work emphasizes on the vital contribution of State Space Models (SSM) in improving detection tasks.