VEHICLE TYPE AND MAKE RECOGNITION BY COMBINED FEATURES AND ROTATION FOREST ENSEMBLE

摘要: Vehicle type/make recognition based on images captured by surveillance cameras is a challenging task in intelligent transportation system and automatic surveillance. In this paper, we comparatively studied two feature extraction methods for image description, i.e. new multiresolution analysis tool called Fast Discrete Curvelet Transform the pyramid histogram of oriented gradients (PHOG). has better directional edge representation abilities than widely used wavelet transform, which particularly appropriate description rich with edges. PHOG represents local shape orientations computed each sub-region, quantized into number bins, thus ascendency its more discriminating information. A composite from can further increase accuracy classification taking their complementary We also investigated applicability Rotation Forest (RF) ensemble method vehicle combined features. The RF contains set base multilayer perceptrons are trained using principal component to rotate original axes features images. class label assigned via majority voting. Experimental results 600 21 makes cars/vans show effectiveness proposed approach. any single model produces performance compared individual neural network classifier. With moderate size 20, ensembles offers rate close 96.5%, exhibiting promising potentials real-life applications.