遮挡情况下的人体检测与跟踪

Human Detection and Tracking under Occlusion

人体检测与跟踪技术在智能视频监控中有很好的应用价值。提出了一种遮挡情况下的人体检测与跟踪算法。人体检测部分,利用RGB颜色模型与均值漂移算法,从混合高斯分割出的含有遮挡人体的目标中提取多个人体头肩模型,计算人体头肩模型的轮廓特征。采用傅里叶描述子与神经网络分类器实现人体检测。人体跟踪部分,采用基于颜色线索的粒子滤波器作为基本的跟踪算法,为了解决遮挡问题,采用三台摄像机实现多视角跟踪人体,利用三台摄像机之间关于主平面的映射关系确定遮挡人体的跟踪结果。该算法提高了人体检测的准确率和人体跟踪的可靠性,可以广泛应用于复杂环境中的多个人体的检测与跟踪。

Human detection and tracking has good application values in intelligent video surveillance. The human detection and tracking algorithm with occlusion are introduced. In human detection part, first, RGB color model and mean-shift are used to calculate the contour features of the head-shoulder model, while get the headshoulder model from objects containing occluded human body getting from GMM. Then Fourier descriptor and neutral network classifier are used to detect human body. In human tracking part, particle filter based on color clue as the basic tracking algorithm is adopted. To solve the occlusion issue, multi-perspective tracking are proposed by using 3 cameras. To determine the result of human tracking by the principal plane mapping relationship between 3 cameras. The algorithm improves the accuracy rate of human detection and reliability of human tracking, which can be used in the complex environment.