人工智能培训

搜索

人工智能论文:跟踪没有铃声和口哨声(Tracking without bells and whistles)

[复制链接]
kuhuhgiu 发表于 2019-3-15 12:38:31 | 显示全部楼层 |阅读模式
kuhuhgiu 2019-3-15 12:38:31 439 0 显示全部楼层
人工智能论文:跟踪没有铃声和口哨声(Tracking without bells and whistles)跟踪视频序列中的多个对象的问题带来了几个具有挑战性的任务。对于逐个检测的跟踪,这些包括对象重新识别,运动预测和处理遮挡等。我们提供一个没有铃声和口哨声的跟踪器,可以完成跟踪而无需特定地针对任何这些任务,特别是我们对跟踪数据进行训练或优化。为此,我们利用对象检测器的边界框回归来预测对象在下一帧中的新位置,从而将检测器转换为Tracktor。我们展示了Tracktor的可扩展性,并通过简单的识别和相机运动补偿扩展了三个多目标跟踪基准,提供了最先进的技术。然后我们对几个状态的性能和故障情况进行分析 - 最先进的跟踪方法和我们的Tracktor。令人惊讶的是,在处理完全跟踪方案时,没有任何专用跟踪方法更好,即小型和遮挡对象或缺少检测。但是,我们的方法解决了大多数简单的跟踪方案。因此,我们将我们的方法作为一种新的跟踪范式进行激励,并指出了有希望的未来研究方向。总的来说,我们表明,一个巧妙利用的探测器可以比任何当前的跟踪方法执行更好的跟踪,并暴露仍未解决的跟踪挑战。
The problem of tracking multiple objects in a video sequence poses severalchallenging tasks.For tracking-by-detection these include, among others,object re-identification, motion prediction and dealing with occlusions.Wepresent a tracker without bells and whistles that accomplishes tracking withoutspecifically targeting any of these tasks, in particular, we perform notraining or optimization on tracking data.To this end, we exploit the boundingbox regression of an object detector to predict an object's new position in thenext frame, thereby converting a detector into a Tracktor.We demonstrate theextensibility of our Tracktor and provide a new state-of-the-art on threemulti-object tracking benchmarks by extending it with a straightforwardre-identification and camera motion compensation.We then perform an analysis on the performance and failure cases of severalstate-of-the-art tracking methods and our Tracktor.Surprisingly, none of thededicated tracking methods are considerably better in dealing with complextracking scenarios, namely, small and occluded objects or missing detections.However, our approach tackles most of the easy tracking scenarios.Therefore,we motivate our approach as a new tracking paradigm and point out promisingfuture research directions.Overall, we show that a cleverly exploited detectorcan perform better tracking than any current tracking method and expose thereal tracking challenges which are still unsolved.人工智能论文:跟踪没有铃声和口哨声(Tracking without bells and whistles) iqwbyWiC0s01CsGa.jpg
URL地址:https://arxiv.org/abs/1903.05625     ----pdf下载地址:https://arxiv.org/pdf/1903.05625    ----人工智能论文:跟踪没有铃声和口哨声(Tracking without bells and whistles)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

kuhuhgiu当前离线
新手上路

查看:439 | 回复:0

快速回复 返回顶部 返回列表