In the modern urban landscape, parking management emerges as a critical challenge and requires intelligent parking solutions, relying on the automation of vehicle detection, classification, and tracking. However, the success of the learning-based solutions depends on the available datasets. While there have been significant strides in creating datasets for object detection and classification in parking lots, a glaring gap persists: the lack of comprehensive datasets tailored for object tracking, especially in dynamic and challenging lighting and partial occlusion conditions. To address this void, we created a new dataset, sourced from live webcam streams of diverse public parking spaces in Romania. This dataset captures parking dynamics across different times of the day and weather conditions, varied lighting scenarios, and a mix of marked and ad-hoc parking situations, which (especially in terms of ad-hoc parking spaces) are not present in existing parking lot datasets. Our motivation is rooted in the ambition to offer a holistic platform that reflects real-world complexities, enabling researchers and practitioners to train, refine and test algorithms that are practically resilient. The dataset consists of 149580 frames stored as 20 video sequences of 5 minutes each, with corresponding annotations. To illustrate some of the challenges this database allows addressing, we include the results of Yolo object detector on various time moments/lighting conditions, which can serve as a baseline to improve the parking spots detection algorithms.