A Fast Visual Map Building Method Based On Optical Flow Algorithm

Recently,people pay more attention to service such as localization service due to the development of technique.Vision-based indoor localization system becomes more and more popular.As the result,vision-based indoor localization system has become main indoor localization method and has been studied and improved.In order to locate the users accurately,we must build a huge image database,Visual Map.It is really laborious and time-consuming to building a visual map of a large area building such as hospitals,train stations,airports and shopping malls because there must be hundreds of thousands of pictures and location information in the database.Therefore,we proposed a fast visual map building method using improved optical flow algorithm.Firstly,this paper gives the purpose and significance of the research and gives the research present situation at home and abroad.This paper also analyzes the development and application of optical flow algorithm,and studies visual indoor localization algorithms such as epipolar geometry and optical flow algorithm.Secondly,because optical flow algorithm is vulnerable to indoor light change,this paper proposes an improved optical flow algorithm based on image gray features to select high gray values pixels and calculate horizontal displacement and vertical displacement.The experimental results show it is reliable and the accuracy of our algorithm is much better.Thirdly,this paper proposes a camera orientation change model to improve optical flow algoritm because orientation change of the camera can lead to lateral deviation of optical flow.This model can calculate linear velocity instead of lateral deviation of optical flow and eliminate it.Finally,this paper use the improved optical flow algorithm to build a visual map and locate the users.The experimental results show the localization error is slight less than traditional method and it is similar to video stre am method,but this method is much more time-saving and convenient than traditional method and the demend of sensors and of the motion of camera is less than video stream method.