A Machine Learning Approach for Reliable Object Tracking

Object tracking is a critical task that finds its applications in various fields including surveillance and autonomous robots. However, most of the work on object tracking has been developed on images and video data. In contrast, the aim of our work is to develop reliable object tracking system based on sequence of measurements which can be obtained from radio sensors, that are more suitable for privacy-concerned applications. In addition, we propose to use linear regression, in contrast to complex data-driven models, to demonstrate its performance against conventional tracking algorithm i.e., particle filter. Our experimental results show that LR can predict a moving object’s position with minimal error and significantly outperforms the particle filter by more than 90%. All the experiments have been validated via simulations.