Human Activity Recognition via Optimized Deep learning with Improved Hierarchy of Skeleton

One of the vast topics is the recognition of human activity that focuses on recognizing a person's particular movement or action based on the sensor data. Due to issues like partial occlusion, background clutter, variations in look, viewpoint, scale, lighting, detecting the human activities from video sequences is a difficult process. A multimodal activity recognition system is required for several applications like human-computer interface, video surveillance systems, and robots for the recognition of human activity. This paper intends to introduce a new human activity recognition model and it involves three process like “(1) Pre-processing, (2) Feature Extraction and (3) Classification”. The pre-processing of input data is done via background subtraction. The pre-processed data are subjected to extract the features, in which an improved hierarchy of skeleton, weighted bag of Visual words, and Local Texton XOR patterns are extracted. Based on the extracted features, the classification process takes place, in which the Optimized Deep Belief Network (DBN) is exploited. For more precise detection, the weight of DBN is optimally tuned via proposed Poor and Rich with Deer Optimization (PRDO) model. Finally, performance of the presented model is computed over the conventional techniques with respect to various performance metrics.