A method of positioning a window of a motor vehicle

Abstract

The invention discloses a method for locating a vehicle window, comprising: acquiring five types of vehicle window images as positive sample images, and simultaneously acquiring negative sample images; calculating five-direction normalized edge intensity statistical features on the images to obtain a feature vector; calculate the feature vectors of all training samples, use the linear SVM in the LIBSVM toolbox to train the sample feature vectors, and obtain two classifiers for positioning the upper right corner of the window and positioning the upper left corner of the window; For license plate recognition, the adjustment image is obtained according to the size of the license plate; for the adjusted image, it is traversed with a step size of 4 pixels to locate the position of the upper right corner of the window and the upper left corner of the window, and the final positioning of the motor vehicle window is obtained by fusion of the left and right positions. The invention has the advantages that the problem of inconsistent size and shape of the car window is solved, the geometric features of the edge of the car window are accurately expressed, the structure is simple, the sample selection is convenient, the training is easy, the precision is high, and the calculation speed is fast.

Description

technical field [0001] The invention relates to the technical field of image processing and pattern recognition in the computer field, in particular to a vehicle window positioning method. Background technique [0002] According to GA / T833-2016, which is the technical specification for automatic recognition of motor vehicle license plate images, the public safety industry standard of the People's Republic of China, the automatic recognition software of motor vehicle license plate images is required in addition to license plate recognition, vehicle brand logo, vehicle type, body color, etc. The main functions also need to cover behavior feature recognition (including: the driver does not wear a seat belt, dials a mobile phone, puts down the sun visor in the main and passenger seats, and pastes a temporary number plate), and appearance feature identification (including: paste the annual inspection mark part, the roof position of the car. , car pendant decoration, etc.), extrac...

AI Technical Summary

Problems solved by technology

Carrying out feature analysis on the entire car window is susceptible to interference from car windows of different sizes and shapes, which increases the amount of calculation and complexity

Second, the design of classifiers is relatively complex, which will affect the positioning effect

Although the neural network with many layers has strong learning ability, the structure of the network is complex and difficult to determine; it may also fall into local extremum during training, resulting in positioning deviation

Images