Circuit Board Thermal Image Registration Algorithm

Circuit card fault diagnosis technique using TIP(thermal image processing) is a kind of non-contacted detection technique. It has such advantages as high speed, high accuracy, low cost, and high rate to detect fault successfully. It can be applied in various fields easily. Besides, the operator need not to know the details of the principle of the circuits.This technique can determine faulty device or faulty area by subtracting the thermal image of a faulty circuit card from that of a standard one. Because of the inevitable instability of the opto-machine scanning of the infrared camera, the location of the same device in the thermal images of the circuit cards, each of which has a different serial number but a same part number , are not exactly same. This phenomena is called 'thermal image excursion'. The accurate detection result can not be obtained unless 'the excursion' has been eliminated.Based on the simple analysis of circuit card fault diagnosis technique using TIP, in this thesis, the registration of circuit card thermal image has been studied, in order to eliminate 'thermal image excursion'. The study consists of the feature point matching, geometrical feature extraction and parameters optimization, etc. Thermal image registration has been realized by the software alone designed in this thesis. Considering the characteristic of the circuit card thermal image, a image registration algorithm based on the feature point has been proposed. In this algorithm, the matching point of each feature point can be obtained by calculating the correlation of the corresponding areas in two thermal images respectively. To enhance the agilities and decrease the calculations, the other thermal image registration algorithm has also been proposed. It is based on the geometrical feature correlation of the thermal image. It chooses the correlation as objective function, and uses the simplex-simulated annealing algorithm to optimize the registration parameters.The experimental results show that the algorithms mentioned above can realize the thermal image registration. Furthermore, the registration precision and display effect can reach a higher level.