基于双流卷积神经网络的智能小车避障算法研究  

The Research of Smart Car Obstacle Avoiding Algorithm Based on Two-stream Convolutional Network

作　　者：张亚初 刘明[1] 赵跃进[1] 董立泉[1] ZHANG Yachu;LIU Ming;ZHAO Yuejin;DONG Liquan(Beijing Institute of Technology, Beijing 100081, China)

机构地区：[1]北京理工大学,北京100081

出　　处：《新技术新工艺》2017年第12期18-23,共6页New Technology & New Process

基　　金：国家自然科学基金(61301190;61475018)

摘　　要：目前,智能小车由于其应用广泛,是研究的热点之一,而基于神经网络与摄像头的智能车由于其在价格以及可靠性等方面的优越性更是受到了广大研究人员的青睐。然而,在端对端的卷积神经网络智能小车避障系统中,有时无论对网络结构和参数进行何种微调,总有一些图像是难以正确分类的,或者说要对某些图像进行正确分类,需要进行相当繁复的参数调整以及结构改变。针对上述问题,考虑采用双流卷积神经网络,利用原始图片与光流图片分别通过预训练的网络后,再将评分结果进行融合,得到最终的分类结果。事实证明,采用该方法可以有效改善传统卷积神经网络一些分类错误的问题。Due to the widely application,smart car becomes a hot research spot today.And many researchers prefer the smart car based on convolutional network and camera because of their advantages in price and reliability.But we found that,in the end-to-end convolutional network obstacle-avoiding system,some images are hard to correctly classification with network structure and parameters adjustment,or need complex adjustment.In this condition,we consider using tow-stream convolutional network,the raw image and optical flow image through the pre-trained network and fuse their scores to get the final classification results.Facts proves that the method can be useful for improving the classification result of traditional networks.

关 键 词：双流卷积神经网络 避障 图像分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]