Abstract
This paper presents a dataset of high-speed recordings of iron ore flowing on a laboratory-scale conveyor belt, captured with top-down videography and organized to highlight both regular operation and the presence of foreign objects. The conveyor belt measures 35 cm in width by 1.10 m in length. It operates at adjustable speeds and is powered by an electric motor to transport hematite and selected contaminants, such as wood pieces or plastic fragments. An NVIDIA Jetson TX2, equipped with its onboard OV5693 camera, recorded the footage at 120 frames per second in 1280 × 720 resolution, using a GStreamer pipeline to stream the video directly to disk. Individual frames were then extracted and sorted into subfolders, distinguishing normal operations from segments containing manually introduced anomalies. Additional subsets further categorize objects by type, enabling adaptation to various detection or classification approaches. This resource is intended to facilitate comparative evaluations of image-based detection approaches in a controlled mining context while also supporting extended uses in computer vision research related to industrial material transportation.