Belt conveyor systems are widely used in industrial settings for bulk material transport, valued for their high capacity and minimal reliance on manual intervention. However, foreign objects on the conveyor can pose significant risks to the system’s operation and downstream equipment. Given the critical role these systems play in maintaining production efficiency, along with their high level of automation and limited opportunities for manual inspection, there is a clear need for continuous, fully automated monitoring, which this paper aims to address. A multi-step solution has been developed which utilizes video stream data from a camera positioned above a conveyor belt. Frames are pre-processed to determine whether the belt is operating, and the edges of the bulk material are detected to establish a region of interest (ROI) for object detection. Each frame is then passed through an object detection pipeline, incorporating several image processing techniques. Edge detection of the ore to establish an ROI is occasionally inconsistent, leading to false detections at the ore edges due to dark thresholding. To mitigate these false detections, a convolutional neural network (CNN) with pre-trained weights from the MobileNetV2 model was implemented as the final step in the object detection process. This object detection system is deployed on an edge device at LKAB Narvik and connected to a SaaS platform that provides predictive maintenance and decision support. The system offers direct insights into whether operations should be halted based on the size and composition of detected objects. Results show a significant reduction in false detections, particularly at the ore edges, and the combination of light and dark thresholding allows for the detection of both high- and low-intensity objects.