摘要
随着智能交通系统、无人驾驶技术以及汽车安全辅助驾驶技术的发展,对在其中扮演重要角色的交通标志识别任务提出了更高的要求。以往的交通标志识别研究会在实时性、准确性以及检测的类别总数上进行一定的改进,但会出现顾此失彼的情况。文章针对这个问题提出了一种改进的Yolov5算法,该算法首先采用自适应锚框计算得到适合实验中交通标志数据集的锚框,并在输入端增加了Mosaic数据增强技术以丰富数据集,然后引入Stem模块以及EfficientNetLite网络结构来替换Yolov5主干网络,最后采用CIoU损失函数来提高预测框检测精度。并在TT100K数据集上进行实验,并将实验结果与原Yolov5以及Yolov5-ShuffleNetV2网络实验结果进行比对。结果表明,提出的方法相较于Yolov5在内存的使用上减少了49.7%;参数量减少了49.9%;模型复杂度降低了59.1%;同时mAP_0.5达到了80.9%,提高了2.1%;检测精确度也提高了2.2%,证明了改进后的算法的有效性。
With the development of intelligent transportation system, unmanned driving technology and vehicle safety assisted driving technology, higher requirements are put forward for the task of traffic sign recognition, which plays an important role in it. Previous research on traffic sign recognition will make certain improvements in real-time, accuracy, and the total number of detected categories,but there will be a situation where one will lose the other. Aiming at this problem, this paper proposes an improved Yolov5 algorithm.The algorithm first uses the adaptive anchor frame to calculate the anchor frame suitable for the traffic sign dataset in the experiment,and adds Mosaic data enhancement technology to the input to enrich the dataset. Then, the Stem module and the EfficientNetLite network structure are introduced to replace the Yolov5 backbone network, and finally the CIoU loss function is used to improve the detection accuracy of the predicted frame. And conduct experiments on the TT100K data set, and compare the experimental results with the original Yolov5 and Yolov5-Shuffle Net V2 network experimental results. The results show that, compared with Yolov5, the proposed method reduces the memory usage by 49.7%;the number of parameters is reduced by 49.9%;the model complexity is reduced by 59.1%;at the same time, mAP_0.5 reaches 80.9%, an increase of 2.1%;the detection accuracy is also improved by 2.2%, which proves the effectiveness of the improved algorithm.
作者
马鹏森
车进
MA Pengsen;CHE Jin(School of Physics and Electronic-Electrical Engineering,Ningxia University,Yinchuan,Ningxia 750021.China;Key Laboratory of Intelligent Sensing for Desert Information,Ningxia University,Yinchuan,Ningxia 750021.China)
出处
《长江信息通信》
2022年第5期10-14,共5页
Changjiang Information & Communications
基金
国家自然科学基金(61861037)。
