基于贝叶斯模型的相机间人群目标识别

准确获取相互遮挡粘连目标的位置特征，是在视野有重叠区域条件下进行相机间目标识别的关键。提出首先构造人体模型，利用贝叶斯模型将粘连目标的分割问题转换为求解最大后验概率问题，然后依据获得的目标轴线特征，在不同的相机间按照最小距离原则进行相同目标的匹配识别。结果表明，利用人体模型进行人群分割的抗干扰能力强，目标识别的准确率较高。

Bayesian Human Recognition Across Multiple Cameras in Crowded Situations

Getting exact human position under occlusion is a key problem to object recognition across multiple cameras with overlapped views.The problem of human segmentation was converted to maximize the posteriori estimation by constructing a human model and a Bayesian model. And then the same objects were matched in different views on the least distance principal by taking the human axis as a feature. Experiments show promising results on human segmentation and recognition in crowded situations and the accuracy rate is high.