Research On Dynamic Recognition Of Spatial Coordinate For Occupant Ears Based On Image Processing Technique

Vehicle interior noise directly affects occupant's physical and mental health and ride comfort.How to control it effectively has been widely concerned by researchers.As a key part in active noise control(ANC)of vehicle interior noise,to track and recognize the spatial coordinates of occupant ears is very important.Based on dynamic image processing technology,a recognition technology for calculating the spatial coordinate of occupant's ear sides in a vehicle is studied in this thesis.Firstly,a measurement system with binocular stereo vision is established.A spatial coordinate is defined according to the requirements of dynamic recognition of the occupant ears sides.The relationship between the pixel and global coordinate systems is investigated.Accordingly,a mathematical model of dynamic recognition system is established.On this basis,the dynamic recognition binocular vision system for spatial coordinate of occupant's ears sides is designed.And the hardware composition and its installation parameters are determined.Secondly,an improved secondary calibration algorithm based on the radial alignment constraint(RAC)method is used to calibrate the dynamic recognition system.Considering the roll,rotation,pitch angles and the main distortion factors of lens,based on the first order radial distortion camera model and RAC calibration method,the internal and external parameters of the binocular camera are obtained.A three-dimensional reconstruction of the ear-side coordinates under real environment is performed and verified by experiments.The experimental results show that the proposed calibration method is reasonable,which can be taken into account in calibration of the dynamic recognition for the occupant ears sides.Thirdly,the areas of ear sides in the images are extracted based on the characteristics of skin color and contour of occupant ear.The histogram model of occupant ear skin color characteristics is established to acquire gray level probability distribution of target imagesand the adaptive threshold segmentation method is used to extract skin color areas.The areas to be detected containing the human ear are extracted through edge contour detection and image morphological operation.A connectivity analysis is conducted to areas to be detected,and characteristic extraction of ear areas is achieved by combining the contour characteristics of human ear.Finally,a stereo matching of the occupant ear areas in the dynamic images is achieved based on the scale-invariant feature transform(SIFT).Through the detection of extreme points in scale space,the localization,direction assignment and generation of eigenvectors of key feature points,the characteristic detection of the areas of occupant ear is performed.The matching dot pairs of occupant's ear areas are obtained based on the K-dimension tree search algorithm and the least squares method,thus stereo matching of human ear images and three-dimensional information acquisition are completed.The Kalman filter is used to track the ear and thus obtain the spatial dynamic coordinates of the occupant's ear area.The experimental results show that the proposed method can realize the dynamic recognition of the spatial coordinates of the occupant's ear sides.This thesis studied a dynamic image-based recognition method for spatial coordinate of occupants' ears sides,which has some theoretical and instructive significance for active control of vehicle interior noise.