Published Paper Details:

In last few years, complex human activity recognition acquired sizeable research attentions. This is due to extensive scope of pattern recognition owing to its noteworthy applications such as robot learning, surveillance, user interface design and so on. For example, activity recognition systems can be applied in human robot interaction in industry to improve their assembly tasks. During the past few years, orientation variations and joint location sequence is widely used for capturing human sequence after extracting discriminative features from frames. However, robust features with complex recognition of human activity were not focused. To address this issue, a Gaussian Polar Coordinate with Sequence Likelihood Support Classifier (GPC-SLSC) framework for complex activity recognition is investigated. To start with, in this work, differentiation between the background and foreground frame are made from raw data using Gaussian Foreground Frame Detection. The obtained features are further processed using Polar Coordinate Feature Extraction model to make them more robust. Finally, the features are trained with Sequence Likelihood Support Vector Machine Classifier for successful complex human activity recognition. The proposed framework is compared with the traditional human activity recognition approaches such as Compressive sensing dictionary-based approach and Long Short Term Memory (LSTM) Recurrent Neural Network where it outperformed them. The results of experiments show that the framework recognizes complex activities involving maximum accuracy with minimum time and complexity