Research On Double-mode Human-robot-interaction Of Small-sample Multi-pose Face Recognition And Gesture Recognition

To realise immersive identity recognition and motion control, this paper reviews thecurrent state of face and gesture recognition. It puts forward a small-sample multi-poseface recognition algorithm along with a depth image and skeleton point based gesturerecognition algorithm, aiming to overcome the drawbacks in multi-pose face recognitionand robust gesture recognition, and to realise multi-mode human-computer interaction.This paper resolves the unstable identity recognition in judicial system caused by thelack of registered samples and therefore leads to broad application prospects. The mainresearch contents are as follows:(1) A novel multi-pose face recognition algorithm that deals with small sample sizehas been introduced. The algorithm first utilises Haar features and random forestclassifier to adaptively locate points of interest that are highly robust to pose variation inthe sample facial images and query images. It then extracts and stores the SURF featuresto generate matching pairs of the pose-invariant points according to their Euclideandistance. The Euclidean distances are further employed to yield an updated matching ofthe SURF feature points and hence complete the face recognition. This method improvesthe accuracy and real-time performance of small-sample multi-pose face recognition.(2) A gesture recognition algorithm using depth image and skeleton points has beenproposed. This algorithm first combines the skin colour model and depth imageprojection to obtain an accurate hand region. It then establishes a3D model of theskeleton points and achieves gesture recognition1D frame of the fingertip distances.(3) An identity recognition system for the small-sample multi-pose framework hasbeen designed. Experiments prove that this identity recognition system overcomes theundesirable issues posed by illumination and shadowing, and the system exhibitsrelatively high robustness and accuracy.(4) A motion control system for the gesture recognition framework which handlesthe motion control of a manipulator of5degrees of freedom has been designed. Theeffectiveness and accuracy of the system is verified.Immersive human-computer interaction systems bring academic significance andbroad application prospects. This paper designs specific face recognition and gesture recognition algorithms that fulfill the need of identity recognition and motion control ascurrent trends. It further implements the algorithms to achieve a highly intelligentmulti-mode human-computer interaction system.