This paper introduces a novel video benchmark dataset for 2D indoor object positioning, focusing on human subjects in industrial environments. The dataset consists of time-stamped video frames synchronized with distance measurements obtained from Ultra-Wideband (UWB) sensors, enabling automated video annotation with object positions. Collected across diverse settings and movement scenarios, it captures realistic challenges such as signal electromagnetic interference and non-line-of-sight conditions. To validate the dataset, we integrated a reference system and developed a GUI tool for analyzing and correcting the collected data. We demonstrate the utility of the dataset by training a machine learning model, which we evaluate against a pure computer vision method, raw UWB-based and reference system’s data. The model has positioning error 10–20 cm. Our cost-effective object positioning solution requires no additional hardware and can be installed onsite within 1–2 days as a basis for precise surveillance.