人员识别_为什么我们需要人员重新识别

人员识别

In this article, I will briefly introduce to you the field of tracking and re-identification without going into technical details. Before talking about re-identification it is essential to mention what person identification (tracking) is and how it works. Tracking is what the security people are responsible for with the only difference being that the machines do all the work. Therefore, a computer receives either some pre-recorded videos from surveillance cameras or real-time video and attempts to differentiate people and classify them. With the help of tracking, we can see the shapes of every person in the scene and identify their movements. This is all great but there are several issues with tracking in real-world scenarios...

在本文中，我将向您简要介绍跟踪和重新标识的领域，而无需涉及技术细节。在谈论重新识别之前，有必要提及什么是人的识别(跟踪)及其工作方式。跟踪是安全人员负责的事情，唯一的区别是机器可以完成所有工作。因此，计算机从监视摄像机接收一些预先录制的视频，或者从实时视频接收它们，并试图区分人员并对其进行分类。借助跟踪，我们可以看到场景中每个人的形状并确定他们的动作。这一切都很好，但是在实际场景中跟踪存在一些问题...

多人跟踪问题 (Multi-person tracking issues)

人员识别_为什么我们需要人员重新识别 — (Image by the author, produced by the Person reID project)

While tracking allows us to receive all the trajectories of movement of anyone in the scene and identify one person from another, the issues start to appear when we have multiple cameras. If the same person moves across a shopping mall and, for example, takes off his jacket in-between cameras, he will not be recognized. Different poses, outfits, backpacks, and other details can mess up our model and recognize the same person as two different ones.

W¯¯往往微不足道跟踪可以让我们接受任何人的运动的所有轨迹场景，并确定从另外一个人，问题开始时，我们有多个摄像头出现。如果同一个人在购物中心购物，例如在相机之间脱下外套，将不会被识别。不同的姿势，服装，背包和其他细节可能会使我们的模型混乱，并将同一个人识别为两个不同的人。

重新识别 (Re-identification)

Re-identification(reID) is the process of associating images or videos of the same person taken from different angles and cameras. The key to the issue is to find features that represent a person. Many of the recent models use deeply learned models to extract features and achieve good performance. Certain state-of-art methods are proposed based on convolutional neural networks (CNN), due to its powerful feature learning abilities and fitting capacity.

R e-identification(reID)是将同一个人从不同角度和照相机拍摄的图像或视频进行关联的过程。问题的关键是找到代表一个人的特征。最近的许多模型都使用深入学习的模型来提取特征并获得良好的性能。由于其强大的特征学习能力和拟合能力，提出了基于卷积神经网络(CNN)的某些最新方法。

reID的良好做法 (Good Practices for reID)

According to the research done by Xiong et al.[3], there are several ways to build an accurate CNN model or improve the existing one. These ways were tested on different re-identification methods and are considered to be generally successful. Let’s take a brief look at these practices since they can help us define what a good re-identification model should look like.

一个 ccording由雄等人所做的研究[3]，有几种方法建立精确模型CNN或改善现有的一个。这些方法已在不同的重新识别方法上进行了测试，并且通常被认为是成功的。让我们简单看一下这些实践，因为它们可以帮助我们定义良好的重新识别模型的外观。

1.在全局池化层之后使用批处理规范化 (1. Using batch normalization after the global pooling layer)

It is a rather technical concept, so in general, we are trying to prevent overfitting during training. Overfitting is when we optimize too much for one dataset and the other examples would become inaccurate because of that.

这是一个相当技术性的概念，因此，总的来说，我们正在尝试防止培训期间的过度拟合。过度拟合是指当我们对一个数据集进行过多优化时，其他示例将因此而变得不准确。

In batch normalization, we aim to normalize the output of each neuron using mini-tach mean and variance. Because of that certain features during training are generalized, which allows us to apply the same model on different datasets.

在批量归一化中，我们旨在使用小数均值和方差归一化每个神经元的输出。因此，训练过程中的某些功能得以概括，这使我们可以将相同的模型应用于不同的数据集。

2.使用一层全连接层进行身份分类 (2. Use one fully-connected layer for identity categorization)

In a CNN there usually are two fully-connected layers. The first one plays the role of a “bottleneck” to reduce feature dimensions. The second layer performs identity categorization.

在CNN中，通常有两个完全连接的层。第一个扮演“瓶颈”的角色，以缩小要素尺寸。第二层执行身份分类。

The authors suggest removing the “bottleneck” layer since it has shown to decrease performance and use the second layer directly. Additionally, it further helps prevent overfitting.

作者建议删除“瓶颈”层，因为它已显示出会降低性能并直接使用第二层。此外，它还有助于防止过度装配。

3.使用Adam进行CCN优化 (3. Use Adam for CCN optimization)

Adam is a rather recently proposed optimization method for the stochastic objective function. Compared to the most popular used SGD, Adam works on lower-order moments, which allows us to smooth the variation between gradients. As you might have thought, it also helps prevent overfitting and alleviate disturbance to the pre-trained model.

亚当是一种针对随机目标函数的最近提出的优化方法。与最受欢迎的SGD相比，Adam可以处理低阶矩，这使我们能够平滑梯度之间的变化。如您所想，它还有助于防止过度拟合并减轻对预训练模型的干扰。

重新识别应用 (Re-identification Applications)

(Video by Accuware Inc.)

(视频由Accuware Inc.提供)

The most typical scenario where it is used is some type of video surveillance. When multiple cameras are located around a shopping mall, parking lot, university, or any other location and we want to ensure security. By using re-identification and tracking models, we are able to follow the path that a person is taking and make sure nothing illegal or inappropriate is done. The only concern for such systems would be privacy, but the surveillance in public places is already functional and active, so adding person tracking would not change the situation.

它是用于吨他最典型的情况是某种类型的视频监控。当购物中心，停车场，大学或任何其他地点周围有多个摄像机时，我们希望确保安全。通过使用重新识别和跟踪模型，我们能够遵循一个人所采取的道路，并确保不做任何非法或不适当的事情。此类系统唯一需要关注的是隐私，但是在公共场所进行的监视已经可以正常运行，因此增加人员跟踪不会改变这种情况。

Additionally, vehicles and other objects could be tracked. In this way, the road situation can be analyzed and further improved.

另外，可以跟踪车辆和其他物体。这样，可以分析并进一步改善道路状况。

结论 (Conclusion)

If the power of computers is used wisely and in a timely manner, crimes and other illegal actions can be prevented and the offenders could be easily tracked. Even though the development of such models is still in progress, their improvement is incredibly impressive and applications are wide.

如果正确，及时地使用计算机的功能，则可以防止犯罪和其他非法行为，并且可以轻松地跟踪违法者。即使这种模型的开发仍在进行中，但它们的改进令人难以置信，而且用途广泛。