A Human Behavior Prediction Method Based on Human Skeleton Motion Information

Abstract

The invention discloses a human body behavior prediction method based on the motion information of the human body skeleton, comprising the following steps: using the human body skeleton information extracted from the RGB-D image to calculate the normalized relative orientation features of each joint point by limbs; The sequence potential difference segmentation method dynamically segments the feature sequence to obtain the gesture feature subsequence and action feature subsequence; extracts key gestures and atomic actions from the gesture feature subsequence and action feature subsequence, and constructs a model based on key gestures and atomic actions. Multi-layer graph model; extract the sub-behavior patterns of the human body contained in the multi-layer graph model, construct the context probability and statistics model of the sub-behavior patterns of the human body; identify and predict the sub-behavior patterns of the human body; It has strong robustness to differences, etc., has strong generalization ability to the action differences of different individuals in the same behavior, and has strong recognition ability to the action similarity between different types of behaviors.

Description

technical field [0001] The invention relates to the technical field of intelligent human-computer interaction and intelligent robot, in particular to a human body behavior prediction method based on human skeleton motion information. Background technique [0002] Human beings will enter an aging society in the 21st century, and the development of service robots can make up for the serious shortage of young labor and solve social problems such as family services and medical services in an aging society. The International Federation of Robotics gives a preliminary definition of service robots: a service robot is a semi-autonomous or fully autonomous robot that can perform services that are beneficial to humans, but does not include equipment engaged in production; if a service robot can be intelligent with humans Friendly interaction, and being able to engage in some household service work according to people's behaviors in daily life, then the application of service robots in...

AI Technical Summary

Problems solved by technology

The International Federation of Robotics gives a preliminary definition of service robots: a service robot is a semi-autonomous or fully autonomous robot that can perform services that are beneficial to humans, but does not include equipment engaged in production; if a service robot can be intelligent with humans Friendly interaction, and being able to engage in some household service work according to people's behaviors in daily life, then the application of service robots in the household service industry will surely form new industries and new markets; in the process of service robots engaging in daily life services , human behavior recognition and prediction is the basis of intelligent interaction between humans and service robots, and plays a vital role in improving the autonomy and intelligence of robots; using human behavior recognition and prediction technology to identify and predict people's daily behavior, It can provide theoretical and practical basis for the daily service of service robots, promote the intelligent and friendly interaction and harmonious coexistence between people and service robots, and improve the comfort of people's lives; there are several problems in the known human behavior recognition and prediction algorithms: first, absolutely Most human behavior recognition algorithms are based on low-level features, that is, training and testing classification models directly based on frame-by-frame pose features, which requires a large amount of calculation; second, when calculating the relative position or orientation of joint points, they are all relative to the body Joint points or hip center joint points do not consider the actual motion model of each joint of the human body, resulting in a more complex motion model for each joint point; third, when extracting key posture features, the entire feature sequence is directly clustered, without distinction The static components and dynamic components of the feature sequence lead to limited accuracy of key poses; Fourth, when segmenting the feature sequence, manual segmentation, fixed number segmentation, fixed interval segmentation, or graph theory-based segmentation are used, and the segmentation effect cannot meet the requirements. ; Fifth, even if the features are extracted by limbs, the behavior model is not modeled by limbs, and the different functions of each limb are not considered. Some algorithms do not consider the coexistence of left-handers and right-handers; Layered Perceptual Model of Behavior

Images