Specific Vehicle Target Detection In The Background Of Massive Images

With the rapid development of Chinese transportation cause,the massive traffic video images are growing each passing day,which brings a huge challenge to the field of image processing.How to effectively storage and process the massive images has become a research hotspot among domestic and foreign scholars at present.Throughout the domestic and foreign researches on massive image processing and vehicle target detection,the existing cloud platform has been used to achieve the extraction of traditional image features,which were directly calculated and matched with the detected vehicle image.However,the features with high dimension and numbers extracted from the massive images not only caused a lot of redundant information,but also produced a great amount of calculation,which will ultimately affect the accuracy and efficiency of vehicle detection.In order to solve the above problems,the thesis studies on the method of specific vehicle target detection of massive images background.The main purpose of the thesis is to efficiently detect the specific vehicle target from the massive images and improve the detection accuracy of the vehicle target,which makes the traffic management and the development of city become more intelligent and efficient.The main contents of the thesis include:1.In order to improve the final detection accuracy and efficiency of the specific vehicle target,the method of extracting the features of the massive images based on Hadoop is studied.According to the characteristics of the vehicle target,the features with every aspect of the vehicle are extracted and described from massive images in the parallel architecture.According to the subjective human vision,human visual feature model is proposed,which includes the extraction of the target area through Harris,subjective proportion feature,subjective global feature,subjective geometric feature of the lights and subjective texture feature of tire.The thesis also considers the objective features SIFT of the target vehicle,which sifts through Harris operator.The above two kinds of features constitute the comprehensive features of the vehicle target.2.In order to improve the final detection efficiency of the specific vehicle target,the high dimensional SIFTs extracted in the thesis need to be dimensionality reduced.According to the characteristics of PCA and LLE,which are respectively used to find the linear structure in high dimensional data and to find the nonlinear manifold structure,PCA is integrated into LLE,and nonlinear reduced dimension algorithm of the local PCA transition is proposed.The method not only eliminates the redundant and irrelevant information in the original high dimensional features,but also saves a lot of unnecessary storage space.3.In order to detect the specific vehicle target,the clustering analysis is used on the features after dimensionality reduction.Bee colony algorithm is combined with the clustering model.The advantages and disadvantages of working mode in team and bee colony clustering algorithm are complementary inspired by the team management,which is formed the parallel intelligent colony clustering in the mode of team management and reduces a lot of calculation for the specific vehicle target detection in the future.The detection efficiency,the calculation of the features,the distance measure and angle measurement,matching measure and structure measure are considered to reconstruct a similarity measurement function.The similarity measure function is calculated to determine the degree of similarity,so as to achieve the purpose of specific vehicle target detection.The simulation results show that the method of specific vehicle target detection of massive images background has good applicability.Compared with the traditional algorithm,the feature extraction method,the improved dimension reduction algorithm and the intelligent clustering analysis are improved.The average detection integrity rate of the test set is 86.38%,and the average detection accuracy rate is 71.25%,which are higher than that of other methods.With closer relationship between road traffic and the development of society and economy than before,the specific vehicle target,which is efficiently and correctly detected in the background of massive images,has the profound significance on all aspects of society and economy.