嵌入粒子滤波中的AdaBoost跟踪器

为了在复杂背景、部分遮挡和光照变化等因素干扰的情况下鲁棒地跟踪视频序列中感兴趣的运动目标,提出了一种新的将分类器融合到跟踪器中的算法。该算法将级联的AdaBoost分类器中每个弱分类器、每层强分类器集成到粒子滤波的观测模型中。同时,结合Fisher线性判别函数在线地选择有鉴别能力的弱分类器用于更好地估计跟踪目标的状态。在不同场景和不同目标的跟踪实验中,提出的算法准确地跟踪到感兴趣的目标。实验结果表明:该算法可以有效地克服复杂背景、部分遮挡和光照变化等因素的干扰,同时可以跟踪目标的快速尺度变化和深度旋转。

AdaBoost tracker embedded in particle filtering

In order to robustly track the interested object in video sequences in the presence of cluttered backgrounds,partial occlusion and illumination variations,a novel approach is proposed to fuse the trained classifier into the visual tracker.The approach integrates all weak and strong classifiers at different stages of the trained cascade using AdaBoost into the observation model for particle filtering.The discriminative weak classifiers are selected online based on the Fisher linear discriminant for better estimates of the tracked object's state.Test results tracking various objects in different scenarios show that the tracker can effectively handle rapid movements,partial occlusion,varying illuminations,and fast changes of scale and view.