摘要
在智能交通管理中,针对如何解决反映各路段全方位交通状况的问题,设计基于无人机航拍的车辆检测系统,该系统对车辆在视频中位置进行检测,并对车辆数量进行统计。以此作为现有交通管理系统的有力辅助手段。首先设计系统的整体框架。接着设计实现基于金字塔L-K光流法的视频稳像,并在NVIDIA Jetson TX1嵌入式平台上实现基于深度学习的车辆检测算法。探索在嵌入式平台上,应用ROS系统框架组织多进程协同运行与通信。该系统实现对于现有交通路段上动态可靠的车辆数量统计和车辆在视频中位置检测功能,对于视频的处理帧率达到14fps。
Aiming at the problem of how to get status of traffic in real time, designs a system which can be used for detecting positions and numbers of vehicles that are shown in airborne video. And it is used in traffic filed as a potent way. First, designs the system＇ s integrated framework. Next, explains in detail that the way of stabilizing video based on pyramidal implementation of the Lucas Kanade feature tracker descrip- tion of the algorithm and realizing algorithm of detecting vehicles based on deep learning on platform of NVIDIA Jetson TX 1. At the same time, it is explored of the method that numbers of progresses and communication among progresses are organized by ROS system. Realizes the functions that counting numbers of vehicles and detecting vehicles＂ positions in the video dynamically and reliably by this system. The frame rate is 14 fps.
作者
韩杰
卿粼波
熊淑华
何小海
HAN Jie QING Lin-bo XIONG Shu-hua HE Xiao-hai(College of Electronics and Information Engineering, Sichuan University, Chengdu 610064)
基金
成都市科技惠民项目(No.2015-HM01-00293-SF)
关键词
无人机航拍
稳像
深度学习
ROS
Aerial Photography
Video Stabilization
Deep Learning
ROS
