基于DSP的交通视频稳像算法设计与实现  

Design and realization of traffic video stabilization based on DSP

作　　者：刘刚[1] 凌强[1] 李峰[1] 王嵩[1] 

机构地区：[1]中国科学技术大学自动化系,安徽合肥230027

出　　处：《微型机与应用》2017年第16期15-18,共4页Microcomputer & Its Applications

基　　金：国家重点研发计划(2016YFC0201003)

摘　　要：针对因交通监控设备抖动而引起的视频图像序列不稳定问题,提出了一种在设备内的DSP上直接工作的快速稳像方法。该方法采用Harris算子提取参考帧图像的特征点,利用金字塔LK光流在当前帧跟踪参考帧的特征点,从而获得运动矢量。通过RANSAC剔除匹配错误或前景物体的特征点对,筛选出全局运动矢量。采用描述复杂运动的单应性模型估算全局运动参数,用单纯形法和模板相似性微调参数优化整体效果。最后进行全局运动补偿,实现图像的稳定。依赖TI公司C674X系列DSP的优化方法,实现了对720×480交通视频精确、实时的图像稳定处理。Aiming at removing the jitter of video sequences caused by traffic monitoring equipment, a fast image stabiDSP in the device is proposed. The feature points of reference image frame are extracted by Harris corner,and the LK optical flow of Pyramid is used to track them in the current frame. The featiare point pairs that are matching error or belong to foreground objects are eliminated by RANSAC. The homography model which is always used for describing the complex motion is used to estimate the gloverall effect is improved by trimming parameters with simplex method and template similarity. Finally,the global motion compensation is per-formed and the stabilization and reconstruction of image are implemented. The accurate and real-time image stabilization system of 720 x 480 traffic video is implemented based on the optimization method of C674X series DSP.

关 键 词：视频稳定 全局运动估计 单应性 运动补偿 DSP 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]