Research On The Key Technologies For Human-computer Interaction Based On The Eye Movements

Some special people with physical disability or paralysis can’t independentlycommunicate with the surroundings by using the traditional human-computerinteraction (HCI) methods which need to use limbs to operate the interaction. Withdevelopments of multiple technologies, new human-computer interaction methodsbased on visual, auditory and olfactory modes are emerging. In this dissertation,some key points of HCI based on visual information are studied in order to helpthose special people communicate with their surroundings independently.The human-computer interaction based on visual information records andrecognizes the different eye motions as the input commands so as to control somedevices. Common eye motions include gaze, saccade (eye moving left, right, up anddown) and blink. All these eye motions are called as eye gestures.This dissertation studies such key technologies as below:1. The technology relating to recording the eye motions is studied. In thisdissertation, the video-oculography（VOG）method is adopted to capture eye imagesthrough the camera and eye motions are identified by image processing. In this way,users won’t be nervous at all during recording eye motions. The system is composedof only one camera and one computer, without any auxiliary light source to produceglints or any image projecting screen to compute the gaze point.To capture clear and complete two-eye images, a mathematical model is built todiscuss the following issues:1) the reasonable region for placing the camera,including the shortest distance from the camera to the user and the maximumdeflection angle of the camera;2) the maximum degree of the head horizontallyrotating. The parameters of the model are simulated and computed by the Matlabsoftware. Then, the Congex Insight1020camera is adopted to capture images of themodel head under the visualable light in our lab. The experimental results show thatthe single camera should locate at one side of the user within±40oso as to capturecomplete two-eye images. And the maximum degree of free head motion is±60o(“±” represents the left/right direction).2. The technology for non-frontal face detection is studied. An improvedAdaboost algorithm is adopted to successfully detect the frontal single face with100%detection rate. Furthermore, the actual head-shoulder images with0~60odeflection angles are sampled under the natural light in the lab. And these images aredetected by using the Adaboost algorithm on the OpenCV platform. The result showsthat we can successfully detect the non-frontal single face with0~±30odeflectionangles.3. The technology for pupil center localization is studied. Eye windows from the detected face images are roughly extracted by using Integral Projection Function(IPF) according to the intensity feature of the eyeball. Based on it, a new method ofMinimum Area Intensity Based on Gaussian Filter (MAIBGF) is proposed tocalculate the coordinates of the pupil center. Then, the Hausdorff universalmeasurement standard is used to test the calculating accuracy. The result shows thatthe proposed method can locate the pupil center from the face images with0~±30odeflection angles with96.43%accuracy.4. The technology for recognizing eye movements is studied. In this paper, therelation between eye movements and changes of pupil center trajectory is analyzed.And the algorithms for horizontal eye movement recognition, vertical eye movementrecognition and the voluntary eye blink recognition are discussed.Finally, experiments on horizontal eye movement tracking, vertical eyemovement tracking, slant eye movement tracking and rectangle eye movementtracking are made under the natural light in the lab. Results prove the crucialtechnologies discussed above are feasible, which provide helpful reference to thepractice of the human-computer interaction based on eye movements.