Research Of Pupil Spatial Position Detection Technology For AR-HUD System

Research of pupil spatial position detection technology for drivers has been a research focus of intelligent vehicle technology.The behavior information of drivers can be obtained by detecting the pupils of drivers and analyzing the trends of pupils' spatial positions,which provide important information for decision systems.The technology has been widely used in many fileds such as human-computer interaction,abnormal behavior detection of drivers,and AR-HUD technology.Pupil spatial position detection technology for drivers can be divided into pupil detection technology for drivers and pupil spatial position measurement technology.In this paper,the pupil detection technology and the pupil spatial position measurement technology are studied in detail,the advantages and disadvantages of various relational algorithms are summarized,and a pupil spatial position detection technology for AR-HUD system is proposed.The innovations of this paper mainly include:1.By analysing the defects of the existing eye detection algorithms in complex scenes with low detection,a human eye detection technology based on cascaded support vector machine along with multi image features and template matching is proposed.The cascaded support vector machine along with multi image features contains two cascaded eye detection classifiers.The first eye detection classifiers is support vector machine along with HOG features.The second eye detection classifiers is support vector machine along with LBP features.The detected eye images of the historical frame is used as the template to locate the eye position of the current frame by template matching while the cascaded support vector machine dose not detect any eyes.The experiment results shows that the proposed method can achieve higher accuracy comparing with the traditional method of single classifier along with single image feature under the condition of real-time control.2.By analysing the limitation of fast radial symmetry algorithm which can only adapt to the situation of circular pupil,which is widely used in precisely pupil location,a pupil center location algorithm based on the improved fast radial symmetry algorithm combining with ellipse fitting is proposed.Firstly,by changing the selection rule of the influence radius of the original fast radial symmetry algorithm,the transformed energy image has more obvious peak.Then,take the peak point as the pseudo-center of thepupil and consider the pseudo center of pupil as the starting point,and fit the edge of pupil by ellipse fitting based on random sample consensus and starburst algorithm.The experiment results shows that the algorithm proposed can not only adapt to the situation of ellipse pupil,but also achieve higher accuracy and better real time performance comparing with the original fast radial symmetry algorithm.3.By analysing the general rules of driver's sitting posture,a viewpoint spatial position measurement algorithm for drivers based on monocular vision is proposed.Firstly,get a single coordinate of the driver's eye position by establishing a constraint model between the driver's eye position coordinates and pupil distance.Then,measure other eye spatial position coordinates by combining the constraint model and the monocular camera model.Take the average value of the spatial coordinates of the two eyes as the driver's viewpoint coordinates.The experiment results shows that the method proposed has high measurement accuracy.4.A smoothing method for a driver's viewpoint spatial coordinates in driving environment is proposed according to the needs of AR-HUD system which requires stable output of the driver's viewpoint coordinates.The driver's viewpoint coordinates are smoothed by the movement model of the viewpoint of the driver established by combining the recursive least squares method with forgetting factor and random consistent sampling to decrease the influence of the eye spatial coordinates' jitter.The experiment results shows that the method proposed can effectively smooth the spatial coordinates of the driver's viewpoint and has low lag.