Human behavior recognition method based on Gaussian process classifier

• Main Authors: WEI JIANMING, WANG XIAOMEI, MA XIAOYUAN
• Format: Patent
• Language: English
• Published: 12.08.2015
• Subjects: CALCULATING; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention provides a human behavior recognition method based on a Gaussian process classifier, which comprises the steps of 1) training a plurality of binary classifier models, wherein each model corresponds to the probability for judging belonging to a certain class of behaviors; 2) acquiring test data of human behaviors, and respectively calculating the probabilities of human behavior classes to which the test data belongs through the Gaussian process binary classifier models; and 3) comparing probability values of the human behavior classes, and determining the human behavior class corresponding to the maximum value to be the human behavior class to which the test data belongs. Compared with a traditional classifier, the Gaussian process classifier can predict the probability of belonging to a certain class of behaviors, various types of weight selectivity can be provided when the class to which the test data belongs is determined. Compared with a classification technique support vector machine which is popular and mature, the Gaussian process classifier has better classification accuracy when the dimensionality of input data is high, and the used training time is obviously shortened when the amount of training data is great.