Research On Driver Ear Recognition And Spatial Localization Based On Vision

Intelligent cockpit is a research hotspot in current automotive technology,and human-computer interaction is the key to it.Human ear recognition is an important part of cockpit active noise reduction and voice interaction,and the accurate and dynamic spatial positioning of the driver’s and occupants’ ears can provide effective support for the human factors design and control algorithm of the acoustic environment of the automotive cockpit,which is one of the important topics in the current research of automotive intelligent cockpit.Therefore,from the perspective of machine vision,this paper collects and constructs the ear data sets of different drivers,proposes an improved Haar-like algorithm to identify the driver’s ear,and then uses the four-point positioning method to achieve the accurate positioning of the ear spatial position.Firstly,pictures of the human ear of different people and angles were collected,and finally more than 1000 ear photos with a size of 40×40 were selected from the positive sample.As much as possible,the feature structure of human ear is included,which provides a sufficient number of sample data for the driver ear training algorithm later.Then,the collected pictures are trained by the improved Haar-like algorithm,and the trained cascade classifier is used to dynamically recognize the driver’s ear in the car.Secondly,for the pictures that have identified the driver’s ear,the depth camera is used to measure the depth distance,the relationship between the depth camera coordinate system and the pixel coordinate system is studied,and after the calibration of the depth camera in the real vehicle,a variety of body postures are changed to convert the spatial three-dimensional position coordinates of the driver’s ear,which is only 1mm worse than the actual measured data.Thirdly,aiming at the failure of the human ear recognition algorithm that may be caused by the occlusion of hair,headphones,masks,etc.,a four-point positioning method is proposed,and the spatial coordinates of the driver’s ear are calculated by face-ear matching.Face recognition is performed on the driver’s facial features,and a three-dimensional point cloud model of the head is established;Through depth camera and coordinate conversion,the spatial relationship between face features and head models is determined,and the driver identity database of four feature points of left eye,right eye,nose and ear is established to realize dynamic positioning from face to ear.A variety of occlusion tests were carried out on the ears of three drivers to verify the accuracy of the four-point positioning algorithm.Finally,the application case analysis of the combination of ear recognition and active noise reduction algorithm was carried out,and the vehicle test was completed.The active noise control algorithm adopts two algorithms: adaptive step Fx LMS algorithm and adaptive step Fx LMS algorithm of particle swarm optimization.The test results prove the effectiveness of the ear recognition and positioning method,as well as its adaptability to the automotive cockpit ANC system.The human ear dataset constructed in this paper can provide effective sample data for automotive developers and machine vision developers.Improved recognition algorithms can improve the efficiency of machine vision recognition and simplify the development process of human ear recognition.In addition,the four-point localization algorithm proposed in this paper can provide a new idea for solving the problem that all previous recognition algorithms cannot be accurately located when the ear occlusion fails,and can promote the development of related theoretical research and engineering applications.