基于多源特征后融合的分层目标检测算法

Multi-layer Object Detection Algorithm Based on Multi-source Feature Late Fusion

目标检测是计算机视觉领域的热门研究课题,是视频内容分析的基础。文中提出了一种基于图像多源特征后融合的分层目标检测算法。在该算法中,使用多级决策的思想对目标检测任务进行粗细两个粒度的划分。在粗粒度层面,先使用HOG特征对图像进行分类,根据分类器的置信度分数,将测试图像分为正例、负例和不确定例。在细粒度层面,使用多种视觉特征以及多种核函数后融合的方法对不确定域中的图像做进一步分类。在同一数据集上设置了3组对比实验。实验结果表明,所提算法在各个评价指标上都有出色的表现,且在实际视频的目标检测中的效果优于Faster-RCNN。

Object detection is a hot topic in computer vision and it is the foundation of video caption.This paper proposed a multi-layer object detection algorithm based on multi-source feature late fusion,and used ways of multi-level decisions to divide the object detection task into two granularities.At the coarse level,the HOG feature was used to classify the images.According to the confidence scores of the classifier,the test images were categorized into positive,negative and uncertain examples.At the fine level,this paper proposed a multi-source feature late fusion method to classify the examples which are in the uncertain field.This paper conducted several comparative experiments on the same data set.Experimental results demonstrate that the proposed algorithm can obtain excellent results in all evaluation metrics,and achieve a better detection result than Faster-RCNN.