基于视频的场景智能分析和人机交互技术研究-计算机应用技术专业论文.docxVIP

11 11 浙江大学硕士学位论文 Abstract Abstract Wi也 the explosive growth of video，people pay more and more attention to the intelligent video ana1yzing and processing. 币le automation of intelligent video ana1yzing and processing wil1 dramatica11y decrease peoples workload. We do research of key a1gorithms in two important regions of intelligent video analyzing and processing: video-based intel1igent scene ana1ysis and video-based human-computer interactlon. After ana1yzing the videobased intel1igent scene 组talysis (the software a1gorithm of intelligent video surveillance)，we extract its key a1gori址uns，and then do research on them: We present a foreground de阳ction a1gorithm based on adaptive dua1-threshold. It solves the problem of gaps and holes produced by many other foreground detection a1gorithms. The foreground point with high-confidence can be detected by 伽 high时 也reshold. Meanwhile ，the lower threshold and mu1ti-level images will have the result th副 can connect the foreground points. In order 阳 track multiobject robust1y，。山tracking a1gorithm adopts two level tracking method，which ωmb讪.es color-histogram based tracking and blob tracking. Firstly the blob trac挝ng is used to do 也e first level tracking. 丁hen 出eωlor-histogram is employed ωdistinguish the objects which cant be distinguished by the blob 衍acking. Employing our foreground detection and object tracking a1gorithm，we a1so develop 由ree key functions in inte1ligent video surveillance: people-counting ，i1legal crossing line detection and removed object de阳ction. Experiment results demons优ate 由副our algori吐uns are robust and practica1， also meet the requirement of application. Moreover，we concem about 由e video-based human-computer interaction and use the posture in video as the input of interaction. We present an a1gorithm based on single camera. Dea1ing with complex background ，we propose a mu1ti伺机le method which is based on veloci