Infant Movements Characteristic Analysis And Research Based On Dense Optical Flow

General movements assessment(GMs), an important method for prediction of infantile developmental disorders at early age, has important implications for establishing training programs and rehabilitation for patient infants, and has been already used in clinic,but the assessment result is result is affected by subjective aspects of the observer assessor's skills need to be maintained for a long time, so it is hard to popularize. This subject is designed to analyzing the characteristic of infant's movements and the rules of infant's general movements based on the computer vision and the dense optical flow,providing more reliable and intuitive data for GMs.Based on the video-based the characteristic of infant's measurement's current research and related learning algorithms, and general movement assessment, this paper studied the babies',whose age are between two and six months, movement features. The main work is as follows:(1) For the video features, this paper puts forward a new method to extract moving object(baby) by color image separation and adaptive threshold. It is a good basis to provide for the subsequent analysis.(2) For analysing local movement characteristic of infants, we divided the baby's body into six parts, the head, the limbs and the trunk, by using watershed segmentation method, and using optical flow tracking each part of the segmentation to achieve coarse segmentation, then combining with the nearest neighbor principle to reclassify the optical flow tracking empty position. So as to achieve fine segmentation to analyse the characteristics of infants' movement by getting each part of the operation center of gravity changes curve and the movement curves of each part of area of the adjacent frames.(3) For analysing global movement characteristic of infants, we superimposed the gray scale images of the baby's body regions on history. Meanwhile, we calculated the optical flow field of baby's body expressing as a color image, getting colorful superimposed image on the history. Then, we also calculated the average velocity and acceleration of each frame of the video sequence, in order to get the changement curvesand mean square deviation of velocity and acceleration. Finally, Statistical analysis of the speed of the infant movement in different directions and the different size of the region, to obtain the statistical histogram, so as to achieve the characteristics of the baby's movement from the perspective of the overall characteristics.(4) Feature extraction based on video sequences of video recording, local and global movement characteristic of infants measurement system interface is complied, including in the Visual Studio 2010 platform, using the C++ and OpenCV.