Font Size: a A A

Research On Key Technology Of Human Behavior Prediction

Posted on:2021-02-09Degree:MasterType:Thesis
Country:ChinaCandidate:K S LiFull Text:PDF
GTID:2428330629982571Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the rapid development of computer science and technology,the degree of artificial intelligence such as machine learning and deep learning has increasingly affected human life.At this stage,the interest in learning computer vision has received widespread attention.Among them,there are many technologies used in life,for example,the human behavior recognition technology in a video is applied in the fields of protection of public security and video surveillance.For the research of human behavior recognition in video,the development of science and technology has improved people 's research methods.For example,the emergence of RGB-D cameras can provide a lot of depth information.At the same time,how to obtain effective information after the information collection is completed is also a big challenge.This is one of the biggest drawbacks of human motion behavior recognition.It has poor real-time performance and a large amount of redundant data.For this shortcoming,behavior prediction can be effectively resolved.The current behavior recognition technology has achieved certain results in human behavior recognition,but the research on behavior prediction is still in its infancy.In 2018,Xidian University proposed a deep learning-based pedestrian gesture recognition and behavior prediction algorithm;in the same year,Xinjiang University published a research paper on the action recognition and behavior prediction methods of a labor-driven lower exoskeleton robot.The results of image edge detection can narrow the range of behavior and action,which is conducive to predicting behavior;the change of human body angle is an obvious feature and easy to collect.Therefore,based on the angle model of key joints of behavioral skeleton,a periodic framed motion prediction algorithm is proposed.First,correctly identify and classify daily behaviors,extract the angle characteristics of the skeleton joint points of the action,set the angle change threshold,and establish the corresponding skeleton model of this type of action;second,extract the unclassified action data frame by frame And the skeletonrefinement process,calculate the angle range of the limbs,through the template matching to predict the action,and determine the key frame of the action.The main innovation of the algorithm to realize the pre-judgment is that it can perform pre-judgment of the motion without completing the recognition of the overall motion,and can determine the key frames of the motion behavior in the video;at the same time,this paper proposes a motion cycle frame Predictive algorithm.First,pre-process the behavior;then,build a skeleton model of human daily behavior based on the action angle feature;then,frame the video and refine the image,and calculate the local overall angular size of the limbs and limbs,and finally,according to the threshold Select the range and predict the next frame of action.
Keywords/Search Tags:Binarization, edge detection, behavior recognition, skeleton refinement, angle change, action prediction
PDF Full Text Request
Related items