Tracking of multiple persons in a wide area by partial information obtained from distributed active cameras

研究了一种通过处理大空间分布式摄像机拍摄的图像来实现多人跟踪的方法，以及通过识别人脸来区分人的方法。主要研究结果如下：1.研究主题1：我们已经开发了一种方法的光流从图像序列中获得的人跟踪。然后，我们开发了一种基于学习阶段中学习到的颜色变化的人位置概率估计方法.研究主题2：我们开发了一种基于人脸候选区域中眼睛和鼻子等人脸部位的人脸检测方法。我们阐明了人脸识别的不确定性取决于人脸在图像中的方向，并建立了识别不确定性的概率模型.研究主题3：我们已经形式化了如何估计一个人在离开现场之前获得的信息。然后，我们定义了评价函数，用于根据迄今为止获得的信息和未来获得的估计信息来确定在哪里查看。研究主题4：我们已经开发了一种方法来确定参数的概率模型的信息在场景中获得的递归模拟.研究主题五：我们构建了一个PC集群系统，并利用PC集群的分布式处理进行了人员真实的实时监控实验。

We studied a method of multiple person tracking by processing images taken from distributed cameras in a wide space, and that of discriminating persons by identifying the face. The main results are the followings:1. Research theme 1: We have developed a method of person tracking by optical flow obtained from an image sequence. Then we have developed a method of probabilistic estimation of person positions based on the color changes learned in a learning phase.2. Research theme 2: We have developed a method of detecting faces based on face parts such as eyes and nose in candidate areas of the face. We clarified that the uncertainty of the face identification depends on the face direction in a obtained image, and developed a probabilistic model of uncertainty of identification.3. Research theme 3: We have formalized how to estimate information of a person obtained before leaving the scene. We then defined the evaluation function, for determining where to look, in terms of information obtained so far and estimated information obtained in future.4. Research theme 4: We have developed a method to determine parameters of the probabilistic model of information obtained in the scene by a recursive simulation.5. Research theme 5: We have constructed a PC cluster system and made experiments of real time monitoring of persons by distributed processing of the PC cluster.