摘要
针对传统方法对实际环境中车辆检测精度不高的问题,提出基于深度学习中R-FCN模型进行车辆检测的方法。基于全卷积网络,结合多尺度训练使模型能够学习到不同尺寸车辆的抽象特征,在训练过程中引入可变形网络提高模型对目标变换的自适应能力,使用软化非极大值抑制的方法减少复杂环境中目标的漏检率。利用Udacity数据集进行训练和测试,实验结果表明,提出方法与R-FCN模型相比,检测的平均准确度提高了4.3%,对实际场景下的车辆有着良好的检测效果,网络具有一定的鲁棒性。
Vehicle objects in real scene,the accuracy of traditional method is not good.To solve this problem,an improved vehicle detection method based on R-FCN model was proposed.The method was based on a fully convolutional network,combined with multi-scale training to enable the model to learn the abstract features of vehicles of different sizes.A deformable network was introduced in the training process to improve the model’s ability to adapt to target transformation.Soft-nms was used to reduce the rate of missed detection of targets in complex environments.The network was trained and tested on the Udacity data set.Experimental results show that compared with the R-FCN model,the proposed method improves the average precision of the test by 4.3%,the vehicle has good detection effects on the vehicle in real scene and the network has certain robustness.
作者
胡辉
曾琛
HU Hui;ZENG Chen(School of Information Engineering,East China Jiaotong University,Nanchang 330013,China)
出处
《计算机工程与设计》
北大核心
2020年第4期1164-1168,共5页
Computer Engineering and Design
基金
江西省自然科学基金项目(20142BAB207001)。
