On application of the histogram of oriented gradients method to vehicles identification

Development of theoretical background, methods and algorithms for identification of visual objects has begun when the first computers appeared. Application of various object recognition techniques in modern technical systems is boosted by fast development of powerful, small and relatively cheap digital devices for data processing that become more and more common these days. In this paper, we study the application of the histogram of oriented gradients (HOG) method to the problem of identification of three kinds of vehicles: cars, planes and ships. The algorithm was implemented in MATLAB and tested using images of Antonov planes and different models of cars and ships. As a result, SVM classifiers for identification of some models of cars, planes and ships were created. To create these classifiers, the authors used sets of images containing the objects to be identified as well as “negative" sets of images that do not contain them. Main parameters of the obtained classifiers were compared. During the simulations, the specialized classifiers for identification of different models of cars, planes and ships were trained and optimal parameters for training and verification were selected to achieve best results. The study showed best results for car objects. To verify the algorithm in real time using real-world images, the authors developed an identification module based on an open-source Orange Pi microcomputer operating under the Android ZIDOO OS.