计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (11): 8-14.DOI: 10.3778/j.issn.1002-8331.1802-0069

• 热点与综述 • 上一篇    下一篇

基于迁移学习和显著性检测的盲道识别

李  林1,李小舜2,吴少智3   

  1. 1.成都师范学院 计算机科学学院,成都 611130
    2.四川大学 电子信息学院,成都 610065
    3.电子科技大学 计算机科学与工程学院,成都 611731
  • 出版日期:2018-06-01 发布日期:2018-06-14

Blind road classification based on transfer learning and salient object detection

LI Lin1, LI Xiaoshun2, WU Shaozhi3   

  1. 1.School of Computer Science, Chengdu Normal University, Chengdu 611130, China
    2.College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China
    3.School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China
  • Online:2018-06-01 Published:2018-06-14

摘要: 为了帮助对视觉障碍患者有效识别道路周围的场景,提出一种基于迁移学习和深度神经网络方法,实现实时盲道场景识别。首先提取盲道障碍物的瓶颈描述子和判别区域集成显著性特征描述子,并进行特征融合,然后训练新的盲道特征表示,用Softmax函数实现盲道场景识别。实验中,对成都不同区域盲道周围障碍物采样,分别采用基于Mobilenet模型不同参数训练和测试了提出的新模型,最后在实际应用场景,实现了盲道周边障碍物的实时分类和报警,实验证明提出的方法具有很高准确率和良好的运行性能。

关键词: 盲道场景识别, 迁移学习, 深度神经网络, 移动网络模型, 显著性检测

Abstract: In order to help the person with visual impairments to detect barriers around blind road, a novelty real-time detection for obstacle scene detection of blind road is proposed. The method is based on transfer learning and feature integration with saliency detection by discriminative region. Firstly, the bottleneck descriptors and saliency map descriptors are extracted through discriminative regional feature integration, and then these descriptors are incorporated together. Secondly, the new model is trained again. Finally, the model is used for real time road scene detection with Softmax function. In the experiments, samples of blind road images are collected at different areas in Chengdu for training and validation with difference model cases. And then, the inference models in real time scene with the embedded device are tested. Experimental results show that the proposed method has advantages on both running time and classification precision.

Key words: blind road scene detection, transfer learning, deep neural networks, Mobilenet, saliency detection