Vehicle Recognition Based On Exponential Moments

Computer vision is fundamentally changing our world, as well as the lifestyle of us. Let the machines replace human eyes, to understand our world,to realize artificial intelligence, and to liberate human hands, which is the dream of countless scientists . The potential of computer vision is infinite,artificial intelligence almost touches all aspects of human life, this paper focuses on the application of computer vision in the field of intelligent transportation research. The author applied exponential moments to vehicle tracking, license-plate recognition, and license-plate character recognition,and finally formed a set of vehicle identification algorithm based on exponential moments.In this paper, exponential moments are taken as an image feature, and a series of studies are carried out on vehicle tracking, license-plate location and license-plate character recognition. The main research work and innovations are as follows:(1) Proposed a new vehicle tracking algorithm based on exponential moments. In the experiment, the author found that at toll-gate and traffic checkpoints, vehicles are driving relatively straight. When observed from a fixed point, the vehicle’s driving away and near could be deemed as the continuous zooming, and this paper utilizes the scaling invariability of the exponential moments to the vehicle tracking, and a new vehicle tracking algorithm is put forward. First, inter-frame difference method is used to confirm the target vehicles. Second, extract the target’s exponential moments,and set it as the tracking parameter. Finally, set the search window to track the vehicle continuously. In comparison with the traditional vehicle tracking algorithm, this paper utilizes the scaling invariability of the exponential moments, lowering the influences from the light and weather, and improving the robustness of the tracking.(2) Proposed a vehicle license-plate recognition algorithm based on exponential moments. In this paper, exponential moments’ translation invariance, scaling invariance and rotation invariance were applied to the vehicle license-plate recognition. It has good recognition effect under the circumstance of license-plate inclined, and good at against weather changes and illumination insufficiency. This method can recognize the vehicle license-plate without tilt correction and proportional adjustment, which not only shorten the positioning time, but also proved the accuracy of the recognition.This research has good practical significance for future smart city contruction.(3) Proposed a new method of license-plate character recognition based on exponential moments and grid feature. The author divides all the vehicles characters into 12 groups, which according to the character morphological characters: the 1st group is a Chinese character group; the 2-11 group is the near modulus group; the 12th group is the modulus uncorrelated group. The 12 groups correspond to 12 neural network classifiers.The characters ready to be processed enter the corresponding classifier. The first judgment is exponential moments, secondly, using grid feature for rejudgment, finally, we get the final recognition results. Because of grid feature, the shortage of exponential moments due to rotation invariance was compensated.