基于YOLO框架的实时交通标志识别算法研究与系统实现

Research and System Implementation of Real-time Traffic Sign Recognition Algorithm Based on YOLO Framework

道路交通标志识别受限于所处复杂的自然场景以及种类繁多等因素,目前少有实时性、准确性、稳定性等各方面均衡的识别系统。利用在工业中应用广泛的YOLO算法对交通标志进行识别,以满足实时性和稳定性;利用大规模的自标数据集对深度网络重新训练,提升其泛化能力,保证识别准确性;将训练完成的深度网络部署在嵌入式设备JetsonTX2上,利用跨平台框架Qt实现交互式界面,使系统具备可应用性。该系统是一个通用平台,可支持不同高精度且高实时的网络模型的替换使用。

Restricted by complex natural scenes and the variety of traffic sign,there are few traffic sign recognition systems that are balanced and practical in terms of real-time,accuracy,adaptability,stability,etc.This paper exploits YOLO algorithm,which is widely used in industry,to recognize traffic signs to satisfy real-time and stability requirements.The deep network is re-trained by using a large-scale self-labeled traffic sign dataset to improve its generalization ability and ensure the recognition accuracy.The trained deep network is deployed on the embedded device Jetson TX2,and the cross-platform framework Qt is used to realize the interactive interface,so that the system is fully applicable.This system is a general-purpose platform that supports the replacement of different high-precision and high-real-time deep network models.