摘要
为了满足目标检测任务实时性的要求,基于轻量级深度学习目标检测网络SSD_Mobilenetv1,通过改进其网络结构,以及增加更细粒特征图参与位置回归和分类来综合网络的上下文信息及引入反残差模块提升网络提取特征的能力,实验表明在保证实时检测速度的同时提高了检测精度,并在KITTI数据集上进行训练验证,取得了良好的效果。
In order to use the video image information to detect and track the target in real time,based on the lightweight deep learning target detection network SSD_Mobilenetv1,by improving its network structure,using the more fine-grained feature map to participate in position regression and classification to integrate the context information of the network and introduce the inverse,the residual module improves the ability of the network to extract features.The experiment shows that the real-time detection speed is guaranteed and the detection accuracy is improved,and the training and verification on KITTI data set have achieved good results.
作者
张俊蓉
徐长彬
唐明周
鹿玮
卞紫阳
ZHANG Jun-rong;XU Chang-bin;TANG Ming-zhou;LU Wei;BIAN Zi-yang(North China Research Institute of Electro-Optics,Beijing 100015,China)
出处
《激光与红外》
CAS
CSCD
北大核心
2019年第8期1019-1025,共7页
Laser & Infrared
