3D Video Motion Perception Model Based On Human Visual System

With the rapid development of the Internet and artificial intelligence,computer vision has a wide range of application scenarios in various industries,and the study of visual perception models in different scenarios has received wide attention from researchers.Currently,the visual mechanism of motion perception in 3D environment is not perfect,and the stereo motion perception models are mostly data models,lacking theoretical support.In this paper,we investigate the problem of 3D video motion perception,explore the characteristics of human perception of stereoscopic motion,and design a physical model of stereoscopic motion perception based on the visual characteristics of the human eye,the details of the research are as follows.To address the problem of limited stereo motion perception mechanisms,this paper designs physiological visual perception experiments to investigate the visual perception mechanisms in human stereo motion scenes.Based on the monocular and binocular visual cues of human eyes,stereo visual stimulus videos of single object motion are designed and viewed by the volunteers to perceive the motion of the target object.The experimental analysis of the factors and perceptual features that influence the volunteers’ correct perception of target motion revealed two findings:firstly,the relative position of the target and the reference affects the volunteers’ perceptual accuracy,and the smaller the relative distance between the target and the reference,the higher the perceptual success rate.Secondly,asymmetry exists in volunteers’ perception of different motion directions of the target,and the perceptual success rate of depth motion can reach 70.13%～73.61%,whichis about 42.67%～47.01%higher than that of lateral motion.However,the perceptual bias caused by errors in depth motion perception is often greater once they occur.In order to solve the problem of motion perception in 3D video scenes,this paper construes the physical model of stereoscopic motion perception based on the visual characteristics of human eyes.First,the model captures the temporal features optical flow vector and spatial features binocular parallax of the target motion in the stereoscopic motion stimulus video,and integrates the spatio-temporal features through spatial transformation to realize the perceptual calculation of stereoscopic motion.To solve the huge perception error of the model for depth motion under the existing features,adding shape features significantly improves the accuracy of the model for depth motion perception.Second,the model exhibits a volunteer-like asymmetric perception capability in the perception task of single-target synthetic video,and its success rate for depth motion perception can reach 46.33%～50.29%,which is about 23.79%-27.79%higher than that of lateral motion,with an error rate of only 0.56%.For the perception of motion components,the model perceives the lateral component of motion more accurately and with lower error than the depth component.The model perceives the motion of a single target in an open environment better in the perception task of real scene video and adapts well.In summary,this paper explores the visual mechanism of the human perception of stereoscopic motion,fills some gaps in the mechanism of stereoscopic motion perception,and clarifies that human perception of stereoscopic motion has asymmetry.Meanwhile,a physical model of stereoscopic motion perception is designed by combining human visual system to solve the motion perception problem of 3D video scenes,and to provide constraints and theoretical support for the data model of stereoscopic motion perception.