改进的YOLOv3算法在道路环境目标检测中的应用

近年来,社会经济持续高速的发展,人均汽车占有量迅速增加。为了避免车辆追尾等事故发生,结合道路环境下目标检测的难点及要求,文章选择基于卷积神经网络的YOLOv3算法,并针对YOLOv3中使用的k-means聚类算法初始时随机选择质心这一不稳定性以及原本的darknet53网络层数较低导致精度不是很高的问题,引用k-means++聚类算法对k-means聚类算法进行优化,并将darknet53替换成特征提取能力更强的resnet101,进行算法优化。实验结果显示优化后的算法mAP提高了12.2%,基本符合实际应用检测的精度要求。

The Application of Improved YOLOv3 Algorithm in Road Environment Target Detection

In recent years,with the rapid development of social economy,the per capita car ownership has increased rapidly.In order to avoid accident,such as rear-end collision,combined with the difficulties and requirements of target detection in road environment,this paper choose YOLOv3 algorithm based on convolution neural network.Aiming at the problem of the accuracy is not very high caused by the instability of K-means clustering algorithm which selects the center of mass in the initial randomly and the low layers of darknet53 network.referencing k-means++clustering algorithm to optimize k-means clustering algorithm,and using resnet101 whose feature extraction capability is stronger to replace darknet53,optimize algorithm.Experimental results show that the optimized algorithm m AP improves 12.2%,which basically meets the accuracy requirements of practical application.