CAJ | 학술논문

视觉监控是计算机视觉研究的前沿方向.动态场景视觉监控就是利用计算机视觉和人工智能的理论和方法,通过对摄像机拍录的图像序列进行自动分析来对场景中的运动物体进行定位、跟踪和识别,并对物体的运动行为作出判断或者解释,达到监控的目的.本文结合交通场景监控这一特定任务,实现一个包括摄像机标定、模型可视化、运动车辆的姿态优化与定位、跟踪预测、基于轨迹分析的行为理解等功能算法的交通场景视觉监控系统.从算法和实现的角度出发,文章对系统中各个功能模块进行了较为详细的描述与讨论.
시각감공시계산궤시각연구적전연방향.동태장경시각감공취시이용계산궤시각화인공지능적이론화방법,통과대섭상궤박록적도상서렬진행자동분석래대장경중적운동물체진행정위、근종화식별,병대물체적운동행위작출판단혹자해석,체도감공적목적.본문결합교통장경감공저일특정임무,실현일개포괄섭상궤표정、모형가시화、운동차량적자태우화여정위、근종예측、기우궤적분석적행위리해등공능산법적교통장경시각감공계통.종산법화실현적각도출발,문장대계통중각개공능모괴진행료교위상세적묘술여토론.
Visual surveillance in dynamic scenes is an active research topics in computer vision. The aim of visual surveillance is to make it possible that the computer can watch or monitor a scene by automatic localization, tracking and recognition of moving objects and semantic interpretation of their behaviors in the watched scene. This paper aims to realize a task-specific traffic surveillance system which consists of modules for camera calibration, model visualization, pose refinement, tracking, trajectory-based semantic interpretation of vehicle′s behaviors, etc. In this paper, we describe each module to give readers a comprehensive view of a visual surveillance system, and also discuss possible further work.