• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Symposium on Multidisciplinary Studies and Innovative Technologies >Recognition of Vehicles, Pedestrians and Traffic Signs Using Convolutional Neural Networks
【24h】

Recognition of Vehicles, Pedestrians and Traffic Signs Using Convolutional Neural Networks

机译：使用卷积神经网络识别车辆，行人和交通标志

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

As the present and future of the robotic world and automation, autonomous vehicles and Advanced Driver Assistance Systems (ADAS) that work in conjunction with autonomous vehicles are important technologies that can benefit drivers through current driving environments. Some elementary factors of these autonomous cars are recognizing surroundings, barriers, pedestrians, traffic signs and other vehicles. In this study, as one of the functions of an autonomous car, the operation of peripheral object recognition is carried out through the use of deep learning which has been mentioned with great accuracy and speed these years in the field of solving problems in machine learning. Signs and objects in various environments, different viewing angles and dimensions can be recognized through the video images taken from the vehicle. Application of object recognition is achieved through the use of 517 images of 10 objects consisting of pedestrians, cars, bicycles and 7 traffic signs, and of convolutional neural networks models including SSD Inception V2, Faster R-CNN Inception V2, Faster R-CNN Resnet 50 and Faster R-CNN Resnet 101, which are known as the basis of deep learning. The models previously trained on the COCO data set are retrained and evaluated on the new data set with the transfer learning method. The new data set is formed by part of the image from the GRAZ-01 and GRAZ-02 data sets and part of the image from the mobile phone camera. As a result of performance analyzes, Faster R-CNN Resnet 101 model is found to be successful in object detection on both images and videos with 85.1% accuracy.
机译：作为机器人世界和自动化世界的现在和未来，自动驾驶汽车和与自动驾驶汽车结合使用的高级驾驶员辅助系统（ADAS）是重要的技术，可以通过当前的驾驶环境使驾驶员受益。这些自动驾驶汽车的一些基本因素是识别周围环境，障碍物，行人，交通标志和其他车辆。在这项研究中，作为自动驾驶汽车的功能之一，外围对象识别的操作是通过使用深度学习来进行的，这些年来，在解决机器学习问题的领域中，深度学习的准确性和速度都很高。通过从车辆拍摄的视频图像，可以识别各种环境，不同视角和尺寸的标牌和物体。对象识别的应用是通过使用由行人，汽车，自行车和7个交通标志组成的10个对象的517张图像以及包括SSD Inception V2，Faster R-CNN Inception V2，Faster R-CNN Resnet在内的卷积神经网络模型来实现的50和Faster R-CNN Resnet 101，它们被称为深度学习的基础。使用转移学习方法对先前在COCO数据集上训练的模型进行重新训练并在新数据集上进行评估。新数据集由来自GRAZ-01和GRAZ-02数据集的部分图像以及来自手机摄像头的部分图像形成。性能分析的结果是，发现Faster R-CNN Resnet 101模型能够以85.1％的准确度成功检测图像和视频中的物体。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号