ConveyorBelt_Monitoring

Belt conveyor misalignment is a common fault in mining and mineral processing,threatening equipment safety and production efficiency. Traditional vision-baseddetection methods suffer from unstable references and poor performance incomplex industrial environments. This work proposes a geometric-constrainedinstance segmentation framework combined with dedicated physical markersfor real-time embedded misalignment monitoring. A distance-aware loss anda boundary-enhanced module are designed to improve geometric measurementprecision in the image (pixel) domain. The system is deployed on an RK3588edge platform with INT8 quantization, achieving an end-to-end throughput ofapproximately 26 FPS and a coefficient of determination of 0.9862 in syntheticdisplacement linearity tests. The method presents a promising visual-computingapproach for industrial conveyor monitoring under harsh conditions; however,the current implementation has been validated only under the evaluated conditions, and physical (millimetre-level) calibration and an independent testset warrant further validation.