A Research On Improved Adaboost Algorithm Based On Real-time Road Multi-view Vehicle Detection

In the information society,people can use mathematical techniques to process and interpret the corresponding mode automatically.The vehicle detection technology is the most significant step in the intelligent transportation system,which covers numerous fields such as pattern recognition,image processing,computer vision,machine learning and probability and statistics,etc.This article mainly aims to study vehicle detection by using Adaboost algorithm of machine learning theory.The main work is as follows:1.A brief overview of the preprocessing operations on the vehicle sample in the early time,dealing with the feature extraction on Haar-like feature descriptors and the integral image establishment,etc.2.After detailed study,we get the declining trend of training error boundaries through the error analysis of Adaboost algorithm.By exploring the improvement of Adaboost algorithm,we proposed a dynamic weighting coefficient AdaBoost algorithm to dispose the problems caused by imbalanced data distribution.Through theoretical and experimental analysis of the convergence of its corresponding improvement algorithm,we can conduct a reasonable interpretation,which improves the stability of the system.3.In order to solve the problem that no information sharing between the levels of each strong classifier cascade of AdaBoost classifiers,we join the level dependent threshold adjustment way into SoftCascade cascade structure to do speed-up training,which,to some degree,can reduce the information waste and improve the recognition accuracy.4.According to the background of vehicle running angles,the vehicle sample is divided into seven angle modes.For each model it trains out a cascade classifier,which,to some extent,reduces the complexity of the vehicle detection mode and the difficulty of pattern learning.5.The vehicle detection algorithm is tested under a large number of test sets that including different weather and illumination conditions.Its result shows that the algorithm has a good effect on the detection of vehicle with a high rate of 94.3%.