Lightweight Road Traffic Cone Detection Algorithm Based on Deep Learning

Road traffic cones are safety isolation signs used for highway and urban road maintenance. Automated equipment can greatly improve operational efficiency and reduce safety hazards for workers caused by traditional manual retrieval methods. Accurate target detection of traffic cones is key to ensuring the normal operation of automated retrieval equipment. To improve recognition efficiency while ensuring model detection accuracy and reducing model computation, a lightweight road traffic cone detection algorithm called TraCone-YOLO (TraCone-You Only Look Once) is proposed in this paper. Firstly, an efficient detection head based on shared depthwise separable convolution is adopted in this paper to make the model more efficient in inference. Secondly, to improve the model's ability to extract multi-scale feature information, the RepC2f module is used in the feature extraction process. Finally, by introducing the ELA (Efficient Local Attention) attention module, an ELA-HSFPN (ELA-Hierarchical Scale based Feature Pyramid Network) network structure is proposed for multi-scale feature fusion, further enhancing the model's feature expression ability.In response to the lack of samples with different weather and light intensities in the original traffic cone dataset called TraCon, data augmentation method is applied to simulate different weather conditions in actual scenarios. The training samples are effectively expanded, thereby improving the model's generalization ability and robustness. The experimental results show that the improved model reduces parameters and computation by 56.2% and 55.6% respectively, with an average precision of 92.6%, which is only 0.4 percentage points lower than the original model. The algorithm proposed in this research can make model lightweight while maintaining good detection performance.