基于局部边缘特征的快速目标检测

Fast Object Detection Using Edge Fragment-Based Features

为了实现复杂环境下形状、尺度变化较大的目标检测,提出一种在复杂背景图片中快速目标检测的算法.该算法采用新的局部边缘匹配特征,通过积分图像技术实现快速计算;通过机器学习算法自动提取样本的局部边缘特征来构建目标模板,且不需要任何手工分割和人工筛选的过程.在UIUC通用图像测试库上的实验结果表明,文中算法可在平移、尺度变化、遮拦和光照变化等条件下快速检测出目标,其精确性与已有算法相当,却大幅提高了实时性.

We present a learning model for object detection in images with complex background. Novel local edge feature with chamfer distance as shape comparison measure are used to form a dictionary of templates. The features can be calculated very quickly using the Integral Image technique. Bagging-Adaboost algorithm is applied to select a discriminative edge features set and combine them to form an object detector. Floating search post optimization procedure is included to remove base classifiers causing higher error rates. The resulting classifier consists of fewer base classifiers yet achieves better generalization performance. Experimental results on UCUI image test sets show that our system can extremely quickly detect objects in varying conditions （translation, scaling, occlusion and illumination） with high detection rates. The results are very competitive with some other published object detection schemes. The speed of detection is much faster than that of existing schemes.