Computer vision aided virtual array based radio frequency tomography : measurement setup and analysis

Abstract

Sixth generation (6G) wireless communication is one of the enablers of joint communication and sensing applications which utilizes a common radio spectrum for both communication and sensing applications. This thesis presents an approach of computer vision (CV) aided virtual array to construct the tomography of an object. A vector network analyser-based measurement system is used to gather channel data continuously for the movement of the object relative to the link, creating a virtual antenna array. The CV-based algorithm extracts the object’s landmarks from the video frames and aligns the landmarks with VNA’s channel data to construct a radio image. A human hand and a cardboard box with hidden wooden blocks were used as the measurement objects to generate the image. The accuracy and the quality of the radio image of the hand and the box are analysed with different measurement times and link distances of 30 cm, 40 cm and 50 cm. The ability of radio frequency (RF) waves to penetrate through the object is examined by images of the hidden object.

The VNA and the camera measurements were synchronized based on the timestamps and suitable interpolation techniques. The shape of the object can be constructed using scattered data points. This study successfully constructs the radio tomography of the hand and the hidden wooden pieces using interpolation techniques such as inverse distance weighting (IDW), radial basis functions (RBF) and Gaussian process regression. The results also demonstrates the capability of radio signals to locate hidden objects, showing its potential applications in security and surveillance. By comparing the images generated at different measurement times and various distances, a 100-second measurement time is found as the optimal measurement time collecting enough data points to generate the image. The image produced at 50 cm show a consistent shape whereas the least distance of 30 cm generated an irregular shape of the object. The findings of this research are useful in future research works and can be scaled to applications in real-world scenarios.