基于快速增量学习的行人检测方法

Fast Incremental Learning Method for Pedestrian Detection

行人检测器如何自适应场景的变化是智能交通的一个难点问题.当离线训练的分类器直接用于特定场景检测行人时,其分类性能将大大降低.针对该问题,提出一种基于快速增量学习的行人检测方法.所提方法的特点是以微小代价通过少量在线样本调整离线级联分类器的参数,同时保留原有分类器的分类能力.首先定义基于级联分类器架构的目标混合损失函数,接着分别对混合损失函数的离线部分和在线部分进行计算,得到离线部分的近似结果,然后对混合损失函数进行优化求解,最终给出快速增量学习方法的算法流程.在行人检测公开数据集上测试,相比于现有的增量学习方法,实验结果表明所提方法可以有效解决行人检测器的场景自适应问题.

Automatic adaptation of a generic pedestrian detector to a specific scene is a difficult problem in intelligent transport system. When an offline detector is applied on a new scene where the testing data does not match with the training data because of variations of viewpoints,illuminations and backgrounds,its accuracy may decrease greatly. To alleviate this problem, this paper proposes a fast incremental learning algorithm for pedestrian detection. The method adjust the parameters of offline cascade classifier using few online samples at a low extra cost, and still retains good generalization ability for common environments. Hybrid objective loss function based cascade classifier is given firstly, which composed of online part and offline part. Then online part and offline part of hybrid loss function are calculated, and meanwhile the offline part is obtained using estimation method. An optimization solution method is proposed to hybrid loss function, and the fast incremental learning algorithm flow is given lastly. Compared to the existing incremental learning methods, experiments demonstrate on pedestrian dataset show that the proposed method can solve the problem of adaptive pedestrian detector.