Acquisition And Analysis Of Visual Motion Information And Interactive Strategy For Human-Robot Interaction

Robotics is one of the most promising research fields currently.With the rapid developments of sensors,electromechanical integration,machine learning,artificial intelligence,novel materials etc.,the robot is upgraded from a tool that can only complete simple and repetitive works,to an equipment with multi-function and intelligence.Various applications of robot can be developed according to the practical requirement.Human-Robot Interaction(HRI)is the emphasis of robotics.The key technologies of HRI includes human perception,action recognition,intention understanding,interaction strategy,etc.,In this thesis,the perception and refinement toward the outliers in the captured human movement data,the action speculation according to incomplete data,and the interaction strategy in the case when actions are not recognized in HRI were discussed.Furthermore,acquisition and analysis of visual motion information and interactive strategy were developed systematically.In this thesis,I proposed a Domestic Service and Home Care-oriented HRI(DSHC-HRI)framework,which categories the interaction between human and robot according to the recognition result of human actions into three interaction modes.(1)The general mode: When the system recognizes the user's action,the robot can respond with the prepared interaction behavior.(2)The mimic mode: When the user poses the specified posture,the robot starts to imitate the user's movement synchronously.And(3)the emotional mode: When the user's action cannot be recognized,the robot can still respond with suitable behavior by analyzing and quantifying the emotion that the human movement expresses.For implementing the acquisition and analysis of visual motion information and interactive strategy,the major works and innovations of this thesis include the following:(1)An offline refinement algorithm for movement trajectory was proposed based on auto-segmentation of motion episodes and relevant vector machine.This algorithm was evaluated by spatial similarity,temporal similarity,and smoothness.The results showed that this algorithm can eliminate the noises and outliers in the original capture data.The refined motion trajectories are closer to the ground truths.(2)A stepwise robust regression algorithm was proposed,which achieved online refinement for movement trajectory.We summarized the generation mechanisms of typical outliers,according to which the stepwise refinement structure was designed.In each step,the outliers were checked and refined by the robust regression algorithm.The experimental results indicated that the effect of this online algorithm is very close to that of the offline algorithm,but its ability of online refinement makes it more useful in HRI.(3)A progressive filtering approach for human action speculation was proposed.The continuous movement was segmented into motion primitives,and their spatial and temporal feature vectors were extracted.Then,an iterative recognition scheme was designed based on locality-sensitive hashing,which can identify the action class during the process of gaining motion primitives.The experiments showed that this approach can reliably speculate the class of the ongoing action at an earlier stage.(4)A movement emotion-based HRI scheme was proposed.Two issues are solved in this scheme.(a)An algorithm of analyzing and quantificating the emotion that the human movement expresses was designed.The main idea was to establish the relationship between the human movement and the emotion based on Laban movement analysis,and then the spatial movement emotion and torso pose emotion were defined and quantified.And(b)the interaction principles of HRI oriented toward home service and care were devised,then the interaction strategy based on movement emotion was constructed.The experimental results indicated that this scheme can analyze the body emotion precisely.The generated interaction behaviors are natural and fluent.The proposed DSHC-HRI framework covers most situations for understanding human movement by the general mode,mimic mode and emotional mode.As a result,the seamless and humanized HRI is achieved basically.