Abstract:
This project introduces an innovative Human Activity Recognition and Multi-Subject
Classification and Monitoring System, leveraging state-of-the-art Deep learning techniques to
monitor human activities and direction without the need for intrusive camera systems. The
system addresses privacy concerns and enhances cost-effectiveness, representing a significant
advancement in surveillance technology. By effectively managing multi-person scenarios, it
surpasses the limitations of previous systems reliant on PIR sensors, particularly in accurately
detecting human activities and classifying multiple subjects. Distinguishing between specific
individuals in densely populated environments, the system aims to positively impact settings
where traditional camera systems compromise human privacy. With the overarching goal of
delivering a versatile and ethically sound surveillance solution, it enhances security measures
while upholding individual privacy rights. This innovative approach represents a paradigm
shift, offering a sophisticated means of observation in diverse human-centric environments.
Additionally, the project proposes a method to detect human direction and activities using fused
information from network-connected PIR sensors, validated through simulations and
experiments, demonstrating improved success rates in real-time detection of human moving
positions and activities. Prior to performing feature enhancement operations on the gathered
data, we first gather and examine the PIR sensor's time-domain signal. We then aggregate the
data based on peak time sequence features. The long short-term memory neural network is then
used to predict the activities and direction of moving subject. Lastly, improved success rates in
detecting the moving activities and direction of humans are used to validate the effectiveness
of the suggested method through simulations and experiments, meeting the real-time detection
needs of multiple IOT-based systems.
