Low-cost visitor counting : Evaluation of Bluetooth and Camera-Based Sensing Methods

Sammanfattning

Visitor counting is important for applications such as retail analytics, crowd monitoring, urban planning, and resource management. However, many existing people-counting systems rely on expensive hardware such as radar, LiDAR, or advanced camera systems, which may limit their practical use for smaller organizations and municipalities. This thesis evaluates the practical suitability, limitations, and trade-offs of low-cost visitor-counting technologies based on Bluetooth Low Energy (BLE) MAC detection and camera-based machine learning for occupancy estimation.The project investigates two low-cost sensing approaches: BLE-based MAC detection using a LilyGO T-Beam ESP32 device and camera-based counting using the YOLOv8n-pose model running locally on a Raspberry Pi connected to a standard webcam. Controlled indoor experiments and limited real-world field studies are conducted to evaluate detection accuracy, practical deployment considerations, and reliability under different environmental conditions. Manual observations are used as reference counts throughout the experiments.The results show that both sensing methods are capable of estimating general occupancy trends using affordable consumer hardware. The camera-based approach achieves lower mean absolute errors (MAE = 2.1) than the BLE-based approach (MAE = 3.9) during the semi-controlled indoor evaluation, indicating closer agreement with the manual reference counts. Both methods also exhibit limitations under certain environmental conditions.The results suggest that low-cost visitor-counting systems may provide useful occupancy estimates for applications such as visitor flow monitoring and general crowd analysis where approximate counts are sufficient. However, given the limited scope of the study, additional validation in larger and more diverse environments is required to determine how well the evaluated approaches generalize to other deployment scenarios.