Research On Vehicle Dimensions Measurement Based On Binocular Vision

In recent years,with the continuous development of the social economy and the automobile industry,the car ownership in China has been rising continuously.At the same time,modern traffic management has been increasingly affected by traffic accidents and early road infrastructure damage caused by vehicle overload.At present,the vehicle overload detection method in our country is mainly overweight detection.And the limit detection method for vehicle outline size is relatively limited,even at the manual measurement level.In order to meet the needs of road traffic safety and intelligent traffic and solve the hidden safety problems caused by the oversize of vehicles,the vehicle classification and automatic measurement technology based on binocular vision is been studied in this paper.The main works are described as follows:(1)Research on manual measurement of vehicle dimensions based on binocular vision.Firstly,the principle of binocular vision is expounded,and then the advantages and disadvantages of different camera calibration methods are analyzed,the camera calibration is carried out by using the Zhang's plane calibration method.When the vehicle is manually measured,a zero-mean cross-correlation stereo matching algorithm is selected according to the characteristics of the vehicle measurement.Laplace's variance is used to judge weak regions in the selected region to solve the problem of large matching errors in the weak texture region.The matching window is enlarged in the weak texture region to improve the matching accuracy,making the manual measurement more adaptable.(2)A vehicle classification method that fuses invariant moment features and HOG features is proposed.The determination of the corresponding measurement parameters by vehicle classification is the key to the automatic measurement of vehicle dimensions.The specific work contents for vehicle classification are as follows: First,the principle of invariant moment features and HOG features and their characteristics are studied.The combination of two features to compensate for a single feature does not fully describe this feature of the vehicle model.Then,the support vector machine is selected as a vehicle classifier model.And the vehicle classification method is validated by collecting vehicle side image production data sets.The experimental results show that compared with single features,the classification effect of multi-feature vehicle classifiers has been further improved.(3)An automatic measurement method for vehicle dimensions based on binocular vision is presented.This measurement method can determine the measurement parameters according to different models,and automatically extract the vehicle characteristics and determine the measurement point for dimension measurement,with the characteristics of rapid and accurate.The main contents of the work are as follows: Firstly,the vehicle foreground and the vehicle body segmentation and extraction are implemented by the background difference method and the HSV color space division method.Then,in order to realize the automatic measurement of the vehicle,it is necessary to determine the measurement point in combination with the characteristics of the vehicle.In this paper,the vehicle's long and high measurement points are determined by using the gray projection algorithm.And the vehicle wheelbase measurement points are detected and positioned by the Hough circle transformation.Finally,according to some known conditions of the vehicle body,the car's cabin rectangle is detected using a constrained Hough linear transformation.Experiments show that the proposed measurement point location method can accurately measure vehicle parameters.