Object Recognition Based On Visible And Infrared Image Fusion

Visible image has abundant detail information, but it can’t display a heat source object in thesmoke obscured or dark environment, infrared image can display the object with blur edge informa-tion in this case. So recognition performance can be improved by fusing infrared and visible image.In this paper, the object recognition method based on simplified contour model matching is re-searched at first and a full range of gesture model library of transports is constructed, then extractingcontour features of transports and removing minor contour by using the minimum perimeter polygonapproximation method, finally, constructing triangles with same bottom and getting three characteris-tic quantities for matching identification. The algorithm has fast speed, but its stability under thechange of perspective and scale should be improved, it can not apply to the recognition of the ob-scured object.In order to overcome the disadvantages above, this paper will go on to research improved SIFTfeature-based visual vocabulary object recognition algorithm. SIFT features have good invariance toscale, rotation, illumination, deformation and small angle changes of images. Structuring visual vo-cabulary statistical histogram feature vector to represent the object, the visual word is SIFT feature,then identifying objects for classification by using k-nearest neighbor rules. Since SIFT features be-long to the local features, this algorithm still has good ability for recognition in the case of the ob-scured object, the stability of object recognition is improved greatly.Since the classical SIFT features are base on the image gray gradient, it is not able to distinguishthe object with the same shape and different colors, Therefore, this paper research on the improvedQ-SIFT features, adding color information into SIFT features by the color interval Matrix, it has ob-vious advantages in the case of light and perspective changes, the advantage in the matching speed isparticularly evident.At last, researching the object recognition based on visible and infrared image fusion on the basisof the classical SIFT algorithm. Studying mainly on the weighted average pixel level image fusionrecognition method and making a comparison experiment on object matching performance, The ex-perimental results show that matching and recognition performance of visible and infrared fusion im-ages are superior to single-mode images.