A nonlinear active learning based on AUC optimization and its application to obstacle detection

Aiming at difficulties in labeling caused by a large number of the samples, as well as uneven distribution of the samples in obstacle detection, a nonlinear active learning algorithm based on AUC (area under the receiver operating characteristic) optimization is proposed. Calculation process of this algorithm is as following. Firstly the AUC optimization method is used to train the nonlinear classifier on the training set. Then all the unlabeled samples are classified with the trained classifier. Secondly all the classified samples are scored using the sample selection function based on AUC optimization, and then the best representative samples are selected according to the scores. Finally these samples are labeled by the expert based on the images and location in the image, and then all the labeled samples are put in the training set. The above process is repeated until the AUC converges. Experiments are performed in outdoor environment image database. Experimental results demonstrate that the proposed algorithm can significantly reduce the workload of labeling the samples, and can solve the problem of the sub-optimal solution caused by the uneven sample distribution. The performance is also better than the other active learning algorithms.