Font Size: a A A

Research On Spatio-Temporal Context Learning Via Superpixels

Posted on:2018-12-24Degree:MasterType:Thesis
Country:ChinaCandidate:C HanFull Text:PDF
GTID:2336330512495425Subject:Procedural Law
Abstract/Summary:PDF Full Text Request
Nowadays,with the establishment of the skynet all-round,images and video has become an important basis of the public security investigation,image processing,video dynamic tracking is also the hot topic in the research of various scholars,the rapid development of artificial intelligence,machine learning,also makes the image video got rapid development.In road surveillance video,often because of some trees,cars or other factors lead to track targets obscured,computer can't complete the continuous tracking,can only rely on manpower to track,this greatly wasted a police force and time,how to keep out or half shade,achieve the goal of fast dynamic tracking,is a focus in the study of this article,is also a research focus in the worldwide.At the same time,in order to solve the problem of fast track,this paper uses the superpixels in the image processing technology segmentation technology,the pixel level into a super problem at pixel level,realized the fast track.As a result of image segmentation quality determines the processing of subsequent problems,so on the choice of super pixel segmentation algorithm has become an important factor.This paper combines the pixel division technology of spatial and temporal context prior model to solve the problem of half a block.The article first introduces the present stage for the development of video target tracking technology both at home and abroad,puts forward the research background and significance,and put forward based on the entropy rate of pixel spatial and temporal context prior segmentation algorithm fusion model to dynamic video target tracking,in practice,usually use low-level features video processing images,and rarely USES middle visual features,this article is to use the middle visual characteristics,build a context model,solving the true figure,get the target position in a new frame of maximum probability.First space-time context prior model is set up,by using the Bayesian framework to do image grayscale characteristics of the target to build,and the correlation with the background,through time and space context model to determine the confidence figure of target location,and in the target area by uniform sampling point tracker,determined to accomplish the goal of the current frame position,.Then,using super pixel segmentation algorithm for segmentation,using the pixel location,distances from the center pixel of calculate the requested target at each pixel location probability,the incredible figure into confidence figure at pixel level,evaluating the position of the target in seek regional each probability problems,the location of the highest probability is a new frame of target location.Finally,using the standard video library in the video tracking,to test algorithm of this paper,the analysis and evaluation is put forward,finally summarizes the outlook.
Keywords/Search Tags:Superpixel Segmentation, Spatio-Temporal Context, Priori Model, Object Tracking, Computer Vision
PDF Full Text Request
Related items