Research On Gesture Instructions Recognition Based On Monocular Vision

Under the background of informationed wars in the future, this paper deals with the commandgesture instructions of unmanned aerial vehicle (UAV) based on monocular vision. In this paper, theproblem is decomposed, and it works on chunks of the program reasonably. The theoretical analysisand experimental research of some key technologies, such as gesture segmentation, feature extractionand identification, are carried out. We select the better algorithm for each process to determine thegesture recognition technology solutions. Based on the analysis of the existing algorithms, this paperputs forward the effective foreground extraction algorithm, dynamic background modeling andsubtraction algorithm, gesture segmentation technology based on circular gradient, imagenormalization algorithm, cascading dynamic gesture recognition thchnology which combines withsupport vector machine (SVM), hidden markov model (HMM) and template matching technology andso on creatively.This article selects the thchnology solutions as follows: Based on the complex dynamicbackground model, it extracts the foreground object containing gesture regions firstly. Then use skindetection and circular gradient operator to extruct complete gesture area from foreground regions. UseSVM to implement static gesture recognition based on Zernike moment characteristics and use HMMto implement trajectory characteristics of dynamic hand gestures recognition. Finally, combine theresult of SVM and HMM to form template vector which is the input of template matching algorithm,to classify dynamic gesture instructions. Experimental results of multiple scenes have shown that thetechnique presented in this paper has high identification accuracy under the complex dynamicbackground and has good robustness and lays a foundation for carrier-based UAV gesture instructionsrecognition technology.