Sparse Representation Based Gesture Recognition And Multi-fingered Hand Interaction

Vision based gesture recognition is consistent with human's natural communication habits and can be long-distance and non-contact interacted,which becomes the popular direction in human-computer interaction research.The effect of recognition is largely dependent on the performance of the recognition algorithm.In recent years,the sparse representation based gesture recognition method is very concerned in pattern recognition because of good robustness and recognition effect.But it still need to be further improved because of the low efficiency,parameter uncertainty and high computational complexity of the sparse solution algorithms.Aiming at the above mentioned problems,the sparse representation classification algorithm is analyzed and improved in detail in the paper.The specific research contents are as follows:(1)The ellipse model is established to segment the gesture in YCb Cr color space and the morphologic processing is used to eliminate the noise point.The Hu invariant moments and HOG feature are extracted and PCA is used to reduce dimension.(2)Aiming at the problem of low precision and parameter uncertainty of OMP algorithm and its improved algorithm,a modified adaptive orthogonal matching pursuit algorithm is proposed.Sparsity estimation and variable step size are introduced in this method to realize sparsity approximation.The experimental results show that the recognition rate of the algorithm is better than other improved OMP algorithms,and the running time is obviously smaller than OMP algorithm and SAMP algorithm,especially in more samples and higher dimension.(3)Aiming at the problem of high computational complexity of the l1-norm based solving algorithm,a l2-norm local sparse representation classification algorithm is proposed.The idea of local sparse representation is introduced in this algorithm,and l2-norm method is used to select the local dictionary.The experimental results show that the algorithm can effectively reduce the calculation time while ensuring the recognition rate,and the performance of the algorithm is slightly better than KNN-SRC algorithm.(4)In order to realize human-computer interaction,firstly,the multi-fingered hand and its control model are established.Then,the gesture recognition interface is designed.Finally,the relevant parameters are set for gesture recognition and the recognition result is converted into control command to control the multi-fingered hand.