Research On Eye Tracking Technology And Its Implementation In Driving Environment

Eye is one of the most important ways for human to obtain information and perceive the outside world.It can obtain about 83% information from the outside world.The fixation state of eyes reflects human visual attention,which can analyze human intention and understand human activities.Eye tracking is a technology to capture eye movement state by extracting eye related information,and then analyze human attention and intention.It is an important clue to infer human attention.In recent years,the rapid development of eye tracking technology has become an important research topic in the field of computer vision.At present,the number of cars is increasing,and the road congestion is becoming more and more serious.Hence,the incidence of traffic accidents is getting higher and higher,resulting in a large number of casualties and huge economic losses.Relevant data shows that drivers in the process of driving inattention is the main reason for traffic accidents.Therefore,driver's eye tracking technology has become a hot research topic in the field of driving safety and driving assistance,which has attracted extensive attention of researchers at home and abroad.Generally speaking,the implementation methods of eye tracking technology are divided into two categories,one is the feature-based method and the other one is the appearance-based method.Compared with the feature-based method,the appearance-based method does not require calibration process before eye tracking,which is more natural and flexible.With the development of deep learning in recent years,the appearance-based eye tracking research in driving environment has attracted more and more researchers' attention.In this paper,the driver's eye tracking technology is studied by using deep learning technology in the application scenario of automobile driving.The main work involves two aspects.Firstly,the desktop eye movement tracking is studied and a desktop eye tracking method based on GoogLeNet is proposed.Secondly,the driver's eye tracking is studied by using deep learning technology,and then,a driving environment-oriented eye tracking system is designed.The main innovations of this paper are as follows:(1)A desktop eye tracking algorithm based on GoogLeNet was proposed.Firstly,Kinectwas used to collect the eye-movement video under the user's desktop environment,and image processing technology was used to obtain the left eye and right eye images.And then,the eye-movement database was established.Then,GoogLeNet was used to design the eye tracking depth model.We fine-tuned the GoogLe Net network,and reduced the parameters of the network by adjusting the size of the convolution kernel of the last average pooling layer in the network.The experimental results showed that this method improved the training accuracy of the network model,and the recognition accuracy of the trained network model was 92.4%.(2)A driver's eye movement tracking algorithm based on multi-channel convolutional neural network was proposed in this paper.We used the ordinary camera to collect the driving video,and then the driver's left eye image,right eye image and face image were extracted.The image data was used to construct the eye movement database in driving environment.Based on multi-channel convolution neural network,this paper put forward a driving environment oriented eye tracking depth model,which used the multi-channel convolution neural network to extract the left eye image features,the right eye image features and the face image features respectively,and integrated the features extracted from the three network channels,and used the classifier to get the classification results.The trained depth network model was robust to both the head offset and the line of sight angle of the driver in driving environment,and finally the tracking accuracy of eye movement was 94.6%.(3)Based on the above research,we developed a driver's eye tracking system in driving environment.The system could detect the key areas of the driver's face and employed the multi-channel convolutional neural network based eye tracking algorithm proposed in this paper to locate the driver's fixation area.At the same time,the system could also record and count the driver's fixation time in different regions during driving,and provided data support for driving behavior analysis,driving assistance and other studies.