基于半监督学习的行人检测方法研究

本文提出了基于半监督学习的行人检测方法,用以解决大量的无标记样本问题。在集成分类器的训练过程中,选择BP神经网络分类器、SVM分类器和KNN分类器作为3个子分类器,利用协同训练机制对各个子分类器进行协同训练。针对半监督学习中误标记样本问题,引入富信息策略和辅助学习策略消除训练过程引入的噪声,同时充分利用无标记样例,进而提高分类器的分类精度。通过对测试集和实时视频进行的行人检测实验,证明了本文方法的可行性和有效性。

Research on pedestrian detection based on semi-supervised learning

In order to implement effective detection and utilize large numbers of unlabeled samples,a pedestrian detection method based on Semi-Supervised learning was presented in this paper.Firstly,BP neural networks classifier,SVM classifier and KNN classifier were selected as the three sub-classifiers,and then,the Co-Training mechanism was adopted to train each classifier.Rich information strategy and assistant learning strategy were added in to remove the wrong-marked samples and improve the accuracy of the algorithm by making the most of unlabeled samples.Through the experiments on the test set and real time videos,the feasibility and effectiveness of the approach are verified well.