改进的卷积神经网络行人检测方法

发布时间：2018-01-18 16:18

本文关键词：改进的卷积神经网络行人检测方法　出处：《计算机应用》2017年06期 　论文类型：期刊论文

【摘要】：为了在行人检测任务中使卷积神经网络(CNN)选择出更优模型并获得定位更准确的检测框,提出一种改进的基于卷积神经网络的行人检测方法。改进主要涉及两个方面:如何决定CNN样本迭代学习次数和如何进行重合窗口的合并。首先,关于CNN样本迭代次序问题,在顺序迭代训练多个CNN分类模型的基础上,提出一种基于校验集正确率及其在迭代系列分类器中展现出的稳定性进行更优模型选择的策略,以使最终选择的分类器推广能力更优。其次,提出了一种不同于非极大值抑制(NMS)的多个精确定位回归框合并机制。精确定位回归框的获取以CNN检测过程输出的粗定位框作为输入。然后,对每个粗定位框应用CNN精确定位过程并获得对应的精确定位回归框。最后,对多个精确定位回归框进行合并,合并过程考虑了每个精确定位回归框的正确概率。更精确地说,最终的合并窗口是基于多个相关的精确定位回归框的概率加权求和方式获得。针对提出的两个改进,在国际上广泛使用的行人检测公共测试数据集ETH上进行了一系列实验。实验结果表明,所提的两个改进方法均能有效地提高系统的检测性能,在相同的测试条件下,融合两个改进的方法相比Fast R-CNN算法检测性能提升了5.06个百分点。
[Abstract]:In order to make the convolutional neural network (CNN) select a better model and obtain a more accurate detection frame in pedestrian detection task. An improved pedestrian detection method based on convolution neural network is proposed. The improvement mainly involves two aspects: how to determine the number of CNN sample iterative learning and how to combine the overlap window. For the iterative order of CNN samples, several CNN classification models are trained by sequential iteration. A better model selection strategy based on the correct rate of check set and its stability in iterative series classifier is proposed in order to improve the generalization ability of the final selected classifier. Secondly. A combination mechanism of multiple precise location regression frames is proposed, which is different from that of non-maximum value suppression (NMSs). The acquisition of accurate location regression frames is based on the coarse positioning boxes output from the CNN detection process. Then. The CNN precise positioning process is applied to each coarse positioning frame and the corresponding precise positioning regression box is obtained. Finally, multiple accurate positioning regression boxes are merged. The merging process takes into account the correct probability of each exact location regression box. The final merge window is based on multiple correlation precise location regression box probability weighted summation. Two improvements are proposed. A series of experiments have been carried out on the common test data set of pedestrian detection in the world. The experimental results show that the proposed two improved methods can effectively improve the detection performance of the system. Under the same test conditions, the performance of the two improved methods is 5.06% higher than that of the Fast R-CNN algorithm.
【作者单位】：江苏省大数据分析技术重点实验室(南京信息工程大学);大气环境与装备技术协同创新中心;
【基金】：国家自然科学基金资助项目(61300163)~~
【分类号】：TP183;TP391.41
【正文快照】： 0引言人是各类社会活动的中心,在各种现实场景中,人都是最为重要的关注对象,如行人过马路、车站安检等。人体目标检测作为一种特定物体检测,是车辆辅助驾驶、智能视频监控和人体行为分析等应用的前提,也可以应用在老年人监护、受害者营救等新兴领域中。随着硬件设备功能的不断

【相似文献】