Various Optimizers' Performances for CNN-Based Hand Gesture Recognition for PDP Assistance

Many elderly people suffer from movement-related problems like Parkinson's Disease (PD). Parkinson ’s Disease Patients [PDP] invariably depend on others to fulfil their daily needs. To convey the need of the Parkinson’s Disease Patient to the caretaker, the Parkinson’s Disease Patient Hand Gesture Recognition [PDP-HGR] system using deep learning is proposed here. So nine simple hand gestures to fulfil the daily need of PDP are formulated and an indigenous HG dataset with 16389 images having simple and complex background was prepared. The PDP was initially informed of the hand gesture and its meaning describing to his daily needs. PDP-HGR system recognize the hand gesture performed by the PDP and inform the need of PDP to the caretaker. Using this dataset, a Convolution Neural Network (CNN) based PDP-HGR system to identify the need of PDP is presented. Five gradient descent-based optimizers viz, Stochastic Gradient Descent (SGD), Adaptive Gradient (Adagrad), Adaptive Delta (Adadelta), Root Mean Square Propagation (RMSProp), Adaptive Momentum (Adam) studied on indigenously prepared Hand Gesture dataset. For PDP, it is found from experimental result that RMSprop outperforms amongst the all optimization algorithm for PDP-HGR system.