Non-cooperative Personnel Tracking with Cross Modality Learning in 5G-enabled Warehouse Application

Asset and personnel visibility is crucial for improving workflow efficiency and reducing waste in smart facility, e.g., warehouse applications. 5G networks and technologies provide the high bandwidth and low latency necessary for communicating and fusing multi-modality sensor data, such as high-definition video, time series with high temporal resolution. In this work, we propose to use cross modality learning to develop a self-learning system for locating and tracking indoor personnel with video and WiFi channel state information (CSI) data. We use video data and our computer vision system to provide location annotation automatically, and train a feedforward neural network model for WiFi CSI data in our localization algorithm. Our experimental results show that our localization system is capable of locating a person with submeter accuracy in real-time without laborious manual data annotation.