Data-driven model for human tracking and prediction using Kalman filter with particle swarm optimization

Abstract

Intelligent monitoring systems have evolved due to technological improvement and innovation in biometric identification technology and protecting lives and property. As a result, intelligent monitoring systems are becoming increasingly prevalent. Consequently, this paper proposes the development of an IoT-based Human Tracking system that incorporates the Kalman Filter (KF) Algorithm. Apart from the deployment of the Kalman Filter, analysis was also done with an optimized Kalman filter with PSO (KF-PSO) using Particle Swarm Optimization. Rather than manually tuning the process noise error, R, and measurement error, Q, which are involved in KF, PSO was also used to adjust the errors to obtain their best values for optimal estimation. As a result, a two-dimensional (2D) Kalman filter is developed. The positions and velocities of the object being tracked (i.e., humans) are estimated in x- and y-directions. The proposed system has been evaluated using ten different human datasets, each consisting of 100 samples. The quality performance of KF and KF-PSO models was also compared using accuracy analysis. In comparison, the KF-PSO model yielded an average Mean-Square error of 17 mm (i.e., 1.7% error), while the conventional KF model gave an average Mean-Square error of 22 mm (i.e., 2.2% error). Hence, the KF-PSO model is a better filter than the conventional KF model because of its higher accuracy.