Artificial Intelligence-Based Human Activity Recognition Using Real-Time Videos
Sumita Gupta, Sapna Gambhir · 2024
In the fields of human behavior analysis, human–computer interaction, and ubiquitous computing, human activity recognition (HAR) has grown significantly. In recent times, approaches based on deep learning (DL) have been effectively used to predict a variety of human actions using time-series data from smartphones and wearable sensors. Time-series data handling remains a barrier for DL-based techniques, even though they do quite well in activity detection. Time-series data still has a few problems, such as difficult feature extraction, highly skewed data, etc. In addition, manual feature engineering is a key component of the majority of HAR techniques. Traditional pattern recognition techniques have achieved significant advancements in recent years. However, the performance of the generalization model may be hampered by the approaches’ heavy reliance on human feature extraction. Deep learning methods are becoming more and more successful, and employing these approaches to understand human behaviors in mobile and wearable computing situations or using vision-based technologies has garnered a lot of interest. Research on HAR using the UCF50 dataset is needed to comprehend human behavior and predict human intentions. ConvLSTM and LRCN, a combination of convolutional neural network (CNN) and long short-term memory (LSTM), are the machine learning (ML) methods we employed in the previous research. The LRCN model gets 92% accuracy when we compare the performance of all the models that were utilized against one another. After getting the results of both models, the main aim is to implement the best resulting model on real-time videos, using a webcam and with improved accuracy, along with multi-person detection .