摘要
行人检测在计算机视觉领域是一个热门的研究方向,在自动驾驶和视频监控等应用中广泛使用。为满足全时段行人实时检测的需求,提出一种基于密集连接的SSD算法。该方法以原始SSD算法为基础网络架构,将模型中的附加特征提取层改为密集跳层连接的结构,并引入特征融合结构来融合红外和可见光图像的特征。在KAIST数据集上的实验结果表明,相比于经典的SSD算法,该算法模型更小、精度更高,且满足实时检测的需求,适合部署在资源受限的移动终端设备上。
Pedestrian detection has achieved more and more attention in the field of computer vision,and is widely used in automatic driving,intelligent video surveillance.In order to meet the needs of pedestrian real-time detection at all times,this paper proposes a SSD based on dense skip connections.The proposed method uses the original SSD as the basic network.The additional feature extraction layers in the network model is changed to the structure of dense skipped connections,and the feature fusion structure is introduced to fuse features extracted from infrared and visible images.Compared with the classic SSD algorithm,the experimental results on the KAIST dataset show that the proposed method takes less memory space,achieves more accurate results and can be used in real-time scenes,which is suitable for deployment on resourceconstrained mobile devices.
出处
《工业控制计算机》
2020年第5期103-104,107,共3页
Industrial Control Computer