Unobstructive human activity recognition: Probabilistic feature extraction with optimized convolutional neural network for classification

• Main Authors: Sivakumar, Kalimuthu, Perumal, Thinagaran, Yaakob, Razali, Marlisah, Erzam
• Format: Conference Proceeding
• Language: English
• Subjects: Algorithms; Artificial neural networks; Deep learning; Feature extraction; Feature recognition; Human activity recognition; Human motion; Literature reviews; Machine learning; Outliers (statistics); Smart buildings
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0179829

Human activity detection is a set of techniques that can be used in a wide range of applications, including smart homes, smart cities, medical health care, and many more. Human Activity Recognition can be able to detected accurately by applying machine learning has been done for more than a decade. It was primarily ducked because of the limitations of computing power at that time and its large computational requirements. Nevertheless, machine learning using a probabilistic approach hasbeen a reinforcementpresently due to the majority of data stemming from the information explosion. Despite the fact that the reported results support the claim of superior results in human activity accuracy, it requires adequate enhancements in a few perspectives to the best of our knowledge, since the activity recognition is related to humans, there are potential outliers in the sensor outcomes. Outliers refer to abnormal attempts at a specific activity. Since the public dataset is available as raw data, there shall be potential outliers that need to be addressed well in the data pre-processing and in further recognition. As the work relies on identifying the abnormal activities of elderly people, the feature extraction needs to be focused more on ’extracting abnormal movements from activity’. It has motivated the research in this specific context to further enhance the activity recognition system for elder people. The objectives of this article aim to gather knowledge on the abnormal attempts of activity, to exploit and enhance the probabilistic feature extraction process in such a way that high-level information activity has to be extracted from the activity sensor status, to make the recognition more precise, the depth features will be extracted with the aid of the deep learning network and to develop an objective model that considers the multiple recognition metrics and the trade-off between them, and to formulate the objective model as an optimization problem, to propose a variant of a learning algorithm in such a way it can optimize the deep learning network to understand the data characteristics by solving the objective model. The remainder of the article is structured as follows: Section 1 is an introduction, Section 2 analyses the literature review or research background work, followed by the features and challenges of existing HAR recognition approaches with tabular form, Section 3 is the research gap or problem statements identified and objectives of research stat