Abstract
This paper introduces a switchable, dual-antenna Wi-Fi tracker that is based on an ESP32-ROOM-DA chip and a BNO085 IMU. The tracker is intended to estimate object orientation in confined spaces by utilizing fingerprinting techniques and differences in RSSI values. The research aimed to provide an alternative that does not necessitate magnetometer calibration or intricate antenna arrays, thereby eliminating the constraints associated with expensive AoA systems and magnetometers that are susceptible to interference. Experiments were conducted in a 5 x 5 m test area of a sports hall, with seven randomly distributed access points (APs) within the sports hall. Five APs were in the LOS (line of sight) zone, and two others, which were available in the building, were in the NLOS (non-line-of-sight) zone. The measurements were performed by a DJI Robomaster S1 robot, which was equipped with the tracker. Training data were collected at 100 points; 14 randomly selected locations were used for testing, with eight distinct orientations for every measurement point. During the measurement, the RSSI from both antennas of individual APs, as well as their SSIDs, was recorded. Additionally, the IMU quaternions were mapped to the cardinal directions (N, W, S, E). Four classifiers were trained using the features gathered: k-NN, SVM, Random Forest, and MLP. The k-NN classifier achieved the best performance (MCC 0.23, F1 score 0.39). The dual-antenna system can distinguish the cardinal directions, as evidenced by the results. However, it is imperative to balance the training dataset and collect a greater number of samples to reduce the effect of multipath and NLOS conditions. With more research, it is possible to use an expanded multi-antenna system and the newest Wi-Fi standards. Furthermore, modifications to the measurement process are planned to guarantee a balanced training set.