Abstract:
In the actual traffic environment, it is highly susceptible to interference from external factors such as illumination intensity and deformation of traffic signs, resulti...Show MoreMetadata
Abstract:
In the actual traffic environment, it is highly susceptible to interference from external factors such as illumination intensity and deformation of traffic signs, resulting in low detection rate. Particularly, traditional neural network only relies on features of top layer for recognition, which leads to a consequence that spatial information and edge pixels are easily lost. Aiming at the disadvantages, we combine the high-resolution pixels of shallow convolutional layers with the semantically strong features of deep convolutional layers, and propose an end-to-end feature pyramid framework to construct the complete semantic information about the target. In the optimization phase, ROI Align, soft-NMS and an improved weighted cross entropy loss function are applied to deal with the problem of the limited target pixels and unbalanced distribution of categories. Experiments on GTSDB dataset illustrate the performance of the proposed algorithm has achieved significant improvements with good generalization and stability.
Date of Conference: 10-12 August 2019
Date Added to IEEE Xplore: 03 October 2019
ISBN Information: