The Research On The Application Of Attribute Learning Based On Image Vision

In recent years, property learning has been a hot topic in computer vision,especially property in learning all aspects of practical application, many scholars have put forward specific applications. In the age-predicted, semantic level description capabilities of traditional identification methods is weak, can not accurately represent the number of feature content; crowding in public places detection, all-weather, all terrain detection algorithm is also not mature enough. Aiming at these problems,combined with the idea of studying the properties improved algorithm is proposed.This paper introduces the background and application of learning attributes, but also describes the common target detection methods, theoretical basis and framework structural properties study, a detailed description of the steps of the method relative attributes of learning and, finally, three technologies sort of learning: Point Pointwise stage method, the level Pairwise method, a listing level Listwise methods, and the three methods were compared.On the basis based on the "Properties learning" method estimates the age of the image analysis, the accumulation of property restoration model proposed to solve the problem of the availability of sparse data, property accumulated as a bridge to connect the low-level features and regression model, reducing sparse of the "semantic gap" effect. Proposed model contains explicit semantic meaning, the calculation is more efficient and does not require additional comments. Experimental results show that compared to other classical methods, property accumulated error is greatly reduced,we can see the advantages accumulate property in terms of the age estimates.Paper presented based on the "Properties learning" public places crowding sequencing model, mainly the study average density and the average speed of the two properties, the extraction density property of the moving object using Gaussian mixture model improved, moving object extraction method using the inter-frame difference speed properties, and then extract the Gist feature, the relative properties learning thought, Ranking SVM algorithm uses the improved crowd, traffic congestion and vehicle mixing of three cases simulated order. The data sequence ascrowding label, public places crowding problem into sorting problem, according to the congestion of the label relative positions to determine the relative degree of congestion, can solve the congestion of the algorithm to a certain extent on the ambiguity problem, experiments and achieved good effect.Finally, we designed and implemented based on the "Properties learning" public places congestion-degree detection system, which first collected data source, and then use the property algorithms to analyze and process traffic information to the statistics of the number of vehicles, through simple tests, the system correct detection rate is higher, but also stable work under harsh environment in general.