Human Action Recognition Based On Principal Component Analysis Of Motion Curves

Human behavior recognition system detects and identifies human behavior in the videoautomatically. In normal circumstances, the computer system uses a camera to monitor anenvironment. The task of identification system is to make large-scale data provided by thecamera to make sense, and identify behavior is important for applications currentlyconsidered. Such a system has a large number of potential applications of human behaviorrecognition becomes a very important area of research. Range of applications from visualsurveillance scenes to innovative human-computer interaction.An example of monitoring applications is monitoring a very busy large public places, suchas the airport terminal or subway stations within suspicious behavior. In these cases, thehuman computer security personnel can assist in monitoring a large area becomes easy.Another monitoring application example is unmanned autonomous fighter, which is rapidlybecoming an important military application. Independent monitoring is an important functionof the UAV.Human-computer interaction, the computer may respond to the operator of a behavior orgestures. An example of a human-computer interaction is the visual interface, the behaviorcan be explained and control a display system of the medical image data. Another example isa video game interface, the players immersed games, play the role of the actual behavior.Human-computer interaction is useful on human robot interaction, gesture-based interfacemakes the interaction between man and robot becomes natural. There is some combination ofthe gesture recognition system for speech recognition to create a multi-mode of theman-machine communication interface. Ability of human-computer interaction, and humanbehavior in the general case of recognition, it is just as important to create intelligent robots inthe field of robotics research. An example is the Honda humanoid robot developed ASMIO[1].As robots become more intelligent, they are also able to understand the environment in whichHuman behavior recognition is an important component of that understanding.In this article, we will distinguish between the behavior and activities to achieverecognition. Human behavior is defined as one or a set of actions that can be identified as thebeginning and end of atomic behavior. An example of human behavior, including extendingthe arms, is the implementation of a gesture or walking step. Human activity is a high-levelevent, and is formed by a group or a series of special human behavior. For example, walkingand running can be considered a basic activity. We also can be extended to include the basicactivities of more advanced activities.This paper presents a new framework for the recognition of human action, it can learn a variety of human behavior and do not have to reduce the ability to identify the complete theidentification of human behavior in a very effective way. Behavior recognition systemsubjects one or more critical points of the body to trajectory. Then recognize the underlyingbehavior by analyzing the trajectory.First, this article established the image database. The database is divided into the motioncapture data sets and human behavior data sets. Motion capture data set contains a walk, run,jump, skip forward, up the stairs, marching six basic human behaviors, trajectory data areexpressed in the form. Human behavior data set includes ten different individuals of lowresolution videos, each perform nine same behaviors. The database is not only provided forthe training sample, but also as a basis for assessing the proposed method.Second, The Gaussian mixture background subtraction technique is improved. In this paper,on the basis of the original background subtraction, modify the color of the pixel distributionset match formula, add brightness threshold R matching formula, and perform backgroundsubtraction in the color model of color space to cut off the shadow. We create new backgroundtraining data. The use of part of the frame to create a new frame, it includes only thebackground contains the pedestrian ears background contains60include only the backgroundwithout the training data training subtracted system. The experimental results show that, usingthe Gaussian mixture background subtraction technique, coupled with our improved, is able toclearly depict the contour of the movement of people in the video in each frame.Finally, this paper designed a framework of human behavior recognition method based onprincipal component analysis of the motion curve. This framework by first tracking the feetand hands in the video, the video is changed to several tracks to handle. We havedemonstrated that the action of one or more key points can be used to identify a variety ofbehavioral. A curve fits each track to smooth out the noise and to produce a continuoussmooth motion curve. And then use the curvature and speed to determine the basic operationof where to begin and end, will results Actions curve segmentation for basic operation. Next,a feature vector is generated for each basic operation. Feature vector contains a fixed numberof dimensions, and having the characteristics of the basic operation, and then this featurevector is mapped to the feature space is formed in the principal component analysis of thetraining data set. By mapping feature vector feature space to begin the classification staged,and this feature space is created using the training data set. Using the K nearest neighborclassification method to calculate the characteristics of the basic operation of the vector mapwith the training data, feature vector mapping between the Euclidean distances, classificationbasic operation. The experimental results show that this method can identify a variety ofhuman behavior with high accuracy.