Recognition Method of Road Traffic Signs Based on Improved SSD Model
In order to assist intelligent vehicles in identifying ahead traffic signs rapidly and enhancing the safety of vehicle travel on roads,addressing the issues of misidentification and omission prone to occur in existing technologies during the recognition process,a new detection model,Rep-SSD(Re-parametrization SSD)is proposed based on optimizing the existing SSD(Single Shot MultiBox Detector)object detection model to ensure real-time performance and recognition accuracy.To enhance the feature extraction capability of the backbone network,the convolutional layers in the SSD backbone network are replaced with re-parametrization convolutional layers.A multi-branch structure is incorporated during training to enhance the network's feature extraction capability,and re-parametrization techniques are employed during inference to simplify the network while maintaining recognition speed.To further enhance the small object recognition capability of the improved network,the CBAM(Convolutional Block Attention Module)attention mechanism is introduced to reinforce the feature information extracted by the network.Experimental results demonstrate that on the German Traffic Sign Recognition Benchmark(GTSRB)dataset,the detection accuracy of the Rep-SSD model reaches 84.7%,with a detection speed of 110 frames per second,indicating it has good detection accuracy and real-time performance.The proposed Rep-SSD model provides a real-time and accurate solution for traffic sign recognition in autonomous driving vehicles.
NETL
NSTL
万方数据
