Using a classifier to track the edge of a conveyor belt

Ensuring that a conveyor belt runs true on an X-ray transmission (XRT) particle sorting system is important for a number of reasons. It limits the damage to the edge of the belt, reduces wear to the conveyor and minimises material spillage. Knowing where the belt is also defines the region of interest (ROI) for the applied algorithm, thus reducing false positives outside the ROI. In addition, correct belt tracking ensures that particles are all contained within the detectable area, avoiding the loss of material. Therefore, ensuring correct conveyor belt tracking can greatly impact the cost and efficiency of the sorting system. This paper presents a novel method for tracking the edge of a conveyor belt in a dual energy X-ray transmission (DE-XRT) sorting system. Using a classifier to determine where the belt edge is, the position of the belt is estimated with a pixel error of less than 15 pixels across all test images. An error of this order is expected as the belt edge is 18 pixels wide in the image. The current method, which is not affected by particles lying against the belt, is shown to be better that the previous approach using morphology, which exhibited poorer tracking accuracy when particles touch the belt. In addition, no additional sensors need to be installed and the belt does not need to be specially manufactured, reducing the cost of the conveyor system.