Key points are not available for this paper at this time.
As the pace of intelligence progresses, research into traffic sign detection algorithms has garnered significant attention. Nevertheless, the current traffic sign detection methods face challenges such as limited environmental information perception and the absence of lightweight models. To address these issues, we introduce a lightweight context-aware traffic sign detection network, leveraging the YOLOV8-n framework. Firstly, we developed a module, termed C2f-DCA-Faster, which effectively reduces the number of network parameters while enhancing the network's capacity to capture spatial location information. This enhancement, in turn, bolsters the network's context-aware abilities. Secondly, we integrated a multi-scale module into the network's neck layer, enabling the extraction of diverse features across various spatial scales. This addition enhances the robustness of information fusion. Lastly, we employed the WIOUS loss function to constrain the regression of target positions, thereby improving the network's generalization of location information and its precision in target positioning. Experimental results on the TTIOOK traffic sign detection dataset reveal that our proposed detection model surpasses the benchmark network's average precision (mAP) by 3.8%. Furthermore, our model boasts a parameter count that is O.199G lower than the benchmark, significantly enhancing the detection accuracy of traffic road signs to meet the demands of real-time detection.
Peng et al. (Fri,) studied this question.