CLASSIFIER TRAINING METHOD AND DEVICE FOR VEHICLE-MOUNTED THERMAL IMAGING PEDESTRIAN DETECTION

摘要

A classifier training method and device for vehicle-mounted thermal imaging pedestrian detection. The classifier training method refers to a method for generating enhanced classifier training samples, and comprises: generating enhanced positive samples on the basis of positive sample labeling information and equalization technology, and analyzing information distribution of non-pedestrian background image blocks using a clustering method, so as to assist in screening enhanced negative samples of different categories; pre-processing the enhanced positive and negative samples by adjusting brightness and boundary information; and clustering the positive samples to obtain a sample scale division criteria for long, medium and near distances, and dividing the pre-processed enhanced positive and negative samples into three training sets, and respectively training three classifiers suitable for classifying pedestrian targets at long, medium and near distances. While ensuring the pedestrian detection accuracy, the present method can reduce the computation overhead of pedestrian detection and enhance the scene adaptability of the classifier. The classifier training device includes an enhanced positive and negative sample generation module, an enhanced positive and negative sample pre-processing module, and a training set division and classifier training module.