Vehicle recognition algorithm based on Haar-like features and improved Adaboost classifier
Le Zhang,Jinsong Wang,Zhiyong An
DOI: https://doi.org/10.1007/s12652-021-03332-4
IF: 3.662
2021-06-18
Journal of Ambient Intelligence and Humanized Computing
Abstract:As the first step of vehicle detection and recognition system, how to quickly and accurately detect the vehicle in a picture is directly related to the subsequent vehicle application research. In order to improve the processing speed of vehicle detection, reduce the false alarm rate of detection, and get better results, the method is applied in real scene, this paper carried out in-depth research on this. Collect traffic and urban road surveillance videos as experimental data, of which 2000 were positive samples and 2000 were negative samples. Firstly, a vehicle image preprocessing is carried out on the collected experimental data, and the image feature is extracted based on gray image and improved AdaBoost algorithm, and then the image enhancement is realized by using multi-scale Retinex. Using this method, we can make the image processing accord with the nonlinear characteristics of the human eye to the brightness response, and avoid the distortion of the image directly processed by Fourier transform. In order to improve AdaBoost classifier, it is necessary to use local binary edge features and train the collected feature samples. In order to highlight the vehicle target and ignore the background, we need to use a selective graying way, which is based on the H component of HSV space. The experimental results show that the accuracy of AdaBoost classifier reaches 85.8%, the recall rate is 80.9%, and the comprehensive performance is very high, which can meet the performance requirements.
computer science, information systems,telecommunications, artificial intelligence