Rearch On Optical Flow Algorithm And The Application In Video Object Detection

Computer Vision technology plays an important role in industrial production and high-tech application nowadays. It mainly concerns remote sensing image analysis, character recognition, medical image processing, multimedia technology, intelligent transportation system, national defence, image database, industrial detection, safety monitoring system, you name it.Video object detection technology is a crucial part of the cmputer vision knowledgy. Its purpose is, with information detected, separating the useful moving pixels from static region automatically. With machine vision detection, people can achieve more efficient production rate, more controllable products quality and implement managing the process database. Classification and choosing step are often integrated in the detection module. As the fundamental parts of the research and application on smart recognition and surveillance system, moving object detection construct the solid base for the subsequent object recognition and tracking. In the modernization course, vision detection goes indispensably without saying. Optical flow field is an important concept in computer vision technology. It has broad application in video object diction and tracking system. Optical flow field represents the motion information of the moving object. It is a two-dimensional (2D) instant velocity field based on a corresponding 2D image, of which the velocity is the projection in a plane from a three dimensional velocity. Optical flow field contributes a lot in moving object detection, image matching and three-dimensional (3D) reconstruction technology.Video object detection technology and optical flow field computation techniques is first presented; following is introduction to research on challenges and application of optical flow field, then two integrated methods are provided for video object detection and segmentation, which improve the efficiency and general performance of the algorithm.This paper first briefly introduces the research work. Then, in chapter 2 and chapter 3, on the basis of previous effort on this part, video object detection methods, optical flow field techniques, as well as their classification, algorithm characteristic and performance comparison, are expatiated in a well-knit way. Chapter 4 lists the challenging problems that optical flow field algorithm is faced with, and makes some points on the application of optical flow techniques. Following the theory of video detection and optical flow computing methods above, two new and innovative integrated algorithms are advanced, of which one is integrating pyramid Lucas & Kanade (LK) method with optical flow feature corner dilating movement template and watershed technique, and another is the combination of optical flow method and temporal difference approach; side-by-side are two corresponding application examples and their experimental results. In the end of the paper, Chapter 6, is a draft summary of the whole text; It also gives the direction of further study.