...
  • 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Multimedia Tools and Applications >On-road vehicle tracking using keypoint-based representation and online co-training
【24h】

On-road vehicle tracking using keypoint-based representation and online co-training

机译：使用基于关键点的表示和在线共同训练进行的道路车辆跟踪

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

This paper addresses the issue of tracking an on-road vehicle from images taken by a camera inside a moving platform. Although existing methods have achieved successful results on tracking general objects such as face and pedestrian, when facing complicated road environments, their performance is unsatisfactory. Therefore, a method specialized for on-road vehicle tracking is needed. The proposed tracking method follows "tracking-by-detection" framework and uses co-training to construct a system with increasing learning ability. The method is mainly composed of tracker, detector, and error collector. The tracker uses keypoint matching to estimate the new location from frame to frame. The detector then fine tunes the location by using templates and knowledge-based methods and outputs the bounding box of the vehicle. At last, the error collector catches all possible errors from tracker and detector, and adds them into a dictionary to avoid similar errors in the future. Due to the error collector, the tracker and the detector can reinforce each other during tracking, thus, we also refer to the detector as online detector. This co-training mechanism leads to an efficient offline detector, which employs integrated information, including classified keypoints, templates, and symmetry, to perform "reappearance detection" when object disappears. The proposed method has been successfully validated by performing experiments with an onboard camera mounted on an on-road vehicle.
机译：本文解决了从移动平台内的摄像机拍摄的图像中跟踪道路车辆的问题。尽管现有方法在跟踪一般对象（例如人脸和行人）方面已经取得了成功的结果，但是当面对复杂的道路环境时，它们的性能仍不能令人满意。因此，需要专门用于道路车辆跟踪的方法。提出的跟踪方法遵循“检测跟踪”框架，并使用协同训练来构建学习能力不断增强的系统。该方法主要由跟踪器，检测器和错误收集器组成。跟踪器使用关键点匹配来逐帧估计新位置。然后，检测器通过使用模板和基于知识的方法对位置进行微调，并输出车辆的边界框。最后，错误收集器从跟踪器和检测器捕获所有可能的错误，并将它们添加到字典中，以避免将来出现类似的错误。由于存在错误收集器，因此跟踪器和检测器可以在跟踪过程中相互增强，因此，我们也将检测器称为在线检测器。这种协同训练机制导致了一个有效的脱机检测器，该检测器使用集成的信息（包括分类的关键点，模板和对称性）在对象消失时执行“重新出现检测”。通过使用安装在公路车辆上的车载摄像头进行实验，已成功验证了所提出的方法。

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