Abstract: An adaptive device and an adaptive method for classifying objects with parallel architecture are provided. The method stores a plurality of scene parameters and a plurality of classifier parameters. The method retrieves at least one image datum and then drags frames to enclose a plurality of obstruction images of the image datum and decides a frame range according to a range of each obstruction image. The method uses a plurality of image processing units to respectively calculate a plurality of obstruction characteristic data of the obstruction images in a one-by-one parallel processing way, and chooses and calculates the scene parameters and the classifier parameters corresponding to the obstruction characteristic data to obtain a plurality of classification data. When the method determines that the classification data belongs to the obstruction classification preset, positions of enclosed obstruction images are outputted to achieve real time detection for obstructions.