| With the development of driving assistance technology in the field of intelligent vehicles,driver real-time pupil detection technology has gradually become a hot research direction.By detecting real-time eye position changes of the driver,important information such as the driver’s line of sight and fatigue state can be obtained,so as to provide key technical support and data for intelligent driving,human-computer interaction and other related systems.The real-time pupil detection technology of the driver firstly obtains the image coordinates of the left and right pupils on the twodimensional face image plane obtained by the camera,and then converts the image coordinates to the three-dimensional coordinates of the eye position in the real space through the camera calibration.This thesis focuses on the pupil position detection technology of the driver’s eyes in the image plane,and proposes a real-time pupil detection method from coarse to fine based on hierarchical search,which effectively improves the accuracy and stability of pupil detection.The main research contents of this thesis are as follows:1.In view of real car driving scenarios,this thesis analyzes the limitations of existing pupil detection algorithms,and proposes a human eye coarse positioning algorithm that combines the position distribution relationship of facial nose-eye structure and HOG feature.Is the basic idea of the algorithm in the human eye detector based on HOG feature detection under the condition of multiple candidate box to the human eye,with the nostrils detector to detect the left and right nostril combined,through the establishment of a human face in the nose-eye structure constraint conditions and the statistical distribution features,candidate box,screen out the unreasonable human eyes keep high confidence level of the human eye candidate box as the output of the human eye rough localization algorithm.Experimental results show that compared with the traditional single feature classifier human eye rough location algorithm,the proposed method has higher detection accuracy under the premise of satisfying real-time requirements.2.In view of the idea that the existing pupil central-precision localization algorithm is usually based on the geometric features of the pupil and can only adapt to the limitations of the circular pupil,this thesis proposes an improved pupil central-precision localization algorithm based on ASEF filter.First improved the training methods of filter from the global scope to local scope,makes the human figure has more obvious local energy peak,and then obtain the pupil center position,and then separately for left and right eye local ASEF filter to the adaptability of the optimal parameters,and through the combination of single and eyes ASEF filter to improve localization algorithm in this thesis,the pupil fine robustness of driver’s posture change.Experimental results show that the proposed method is more robust to the pose change of the driver’s head than the pupil precision localization algorithm based on a single filter,thus improving the detection accuracy.3.In this thesis,the integrated use of rough positioning to the pupil fine positioning by human eyes pupil detection method based on hierarchical search,designed the driver real-time pupil detection system hardware and software platform,and applied to real vehicle AR-HUD projects,completed the real vehicle driving experiments,the proposed real-time pupil detection system is verified in computing ability is weak enough to use on-board embedded system has a good performance. |
