Vision Based Navigation For Capsule Endoscopies

Since the world’s first capsule endoscopy "M2A" was successfully developed inIsrael in1999, there have been several products available. Compared withtraditional ones, capsule endoscopy effectively eases patients’ pain in the process ofdiagnosis and extended the range of detection into small intestine. For anautonomous set, it’s essential to equip with a navigation system. Allowing forminimizing the size of capsules, we use the on-broad camera to realize a visionbased navigation for capsule endoscopies.The algorithm based on dark regions was first improved by setting the regionof interest (ROI) to limit the search coverage around last navigation point,introducing the morphology method to eliminate diverticula and vesicae, raising theconcept of position weight to reckon navigation points in the case of lacking forvalid dark regions. Besides that, we proposed the method based on templatematching, which is common in industrial detection, to navigate when the methodmentioned above fails. The two methods were fused and it’s proved beneficial toboth the success rate and the computation speed.Simulation and experiment were carried out to test the effectiveness of theproposed algorithm. The simulation based on a wealth of endoscopic images provedthat the algorithm was better than the existing methods. In the experiment, thealgorithm was applied into an intelligent vehicle. It is proved that the vehicle couldgo through the hose smoothly and the speed met the requirement of real-timenavigation.