Research On The Evaluation Algorithm Of Infant Sleep Quality Based On Computer Vision

Sleep is a crucial part of infants’ life,and having good sleep is beneficial to the improvement of their intelligence and physical development.Aiming at the common sleep problems of infants such as difficulty falling asleep,frequent night awakenings,and disturbed sleep rhythm,this thesis proposes a computer vision-based infant sleep quality assessment algorithm.By establishing an infant sleep detection model and researching infant sleep quality assessment methods,it breaks through the current non-contact lowaccuracy sleep detection technology,reduces hardware costs,and provides parents with a real-time,efficient and convenient sleep detection method.The specific work of this thesis is as follows:（1）Aiming at the complex background of the actual surveillance video,this thesis builds a model that combines eye state recognition and optical flow motion detection.When the target detection cannot judge the baby’s sleep state,the optical flow motion detection is used to assist in judging the baby’s sleep state.（2）In order to improve the rate of the sleep detection model,this thesis adopts a lightweight structure design method and designs a new lightweight network based on Yolov4.In order to improve the rate of the sleep detection model,this thesis adopts a lightweight structure design method and designs a new lightweight network based on Yolov4.On the premise of ensuring the accuracy of the model,the detection processing speed of the model data is improved,and the memory consumption caused by the model itself is reduced.In addition,in order to further improve the generalization ability of the model,this thesis also considers the transfer learning method,thereby improving the performance of the model.（3）Aiming at the problem that optical flow motion detection is easily affected by noise,this thesis compares two optical flow estimation methods.One is based on intrinsic image decomposition and the other is based on four-frame difference.After experimental analysis,the latter has better denoising effect than the former,so the second method is finally adopted.And this thesis uses the projection method to perform deeper denoising when estimating the amount of motion;（4）According to the actual monitoring video,this thesis presents a sleep quality assessment method.Infants of different ages have different sleep habits and sleep structures.In this thesis,the sleep quality assessment model is obtained by using the nighttime sleep data of infants and combining the optimal sleep time of infants of different ages.The feasibility of this thesis is verified by comparing with the scores of professional doctors.