Reminder Detection On Stationary Component Image

Limited to the present manufacture technology and environment, during the production process of industrial component, inevitably some reminders will adhere to the surface of the component, like dross and chipping. Currently most studies on reminder detection have focused on PIND(particle impact noise detection) method. Supported by the theory of human vision, a new method for reminder auto-detection is presented based on image registration with image significant geometric feature.This paper discusses the following contents: (1) the reminder detection system model, which includes seven levels: image input level, image classification level, image geometry extraction level, feature extraction level, geometric primitives/regional matching level, image registration/correction level and reminders detection level; (2) the design of reminder detection system, which includes two parts, the design of image acquisition hardware platform and the design of software algorithms detection platform; (3) three typical detection mode. For three typical element models: non-metallic rectangular element, non-metallic circular element, and a significant single-line element, this paper discuss three kinds of reminders detection mode: round model, remarkable single-line mode and multi-line mode. With seven level of the reminder detection system, each detection mode includes the following algorithm: remarkable geometric primitive extraction algorithm, image registration algorithm and reminders detection algorithm. This paper mainly discuss image registration algorithm based on a significant line primitives, image registration algorithm based on a collection of line, and registration parameter calibration algorithm.We have 150 tests on non-metallic quadrilateral element, 150 tests on non-metallic circular element, as well as 900 tests on a significant single-line element pictures. The correct rate is more than 90% and the speed can reach 1 frame/sec, which meet the system functional requirements and performance requirements.