Abstract: Concerning the problem of tracking interested moving target inaccurately because of complex background, a robust tracking algorithm based on adaptive multi-feature fusion was proposed. First, the algorithm obtained the weighted color distribution model of interested moving target in the HSV color space. Then invariant moment was used to eliminate the interference of the similar background color and illumination changes. The algorithm fused the two features in the particle filter by adjusting their weights and updating particle weights adaptively. Thus, the algorithm can track the moving target accurately and stably. The experimental results show that the algorithm can track interested moving target accurately when moving target is tracked under complex background such as translation, variant posture, and be blocked of the moving object, varying illumination and the interference of the similar background color. The algorithm has strong robustness to background interference.

Key words: weighted color distribution, invariant moment, adaptive fusion, particle weight, particle filter

摘要： 针对由于复杂背景的干扰而导致不能准确跟踪感兴趣运动目标的问题，提出一种基于多特征自适应融合的粒子滤波跟踪算法。首先在HSV颜色空间中得到感兴趣运动目标的加权颜色分布模型，然后利用不变矩特征来消除背景中相似颜色物体和光照变化的干扰，两种特征通过自适应调整权重来更新粒子权值而融合于粒子滤波算法中，从而能够准确和稳定地跟踪运动目标。实验证明，该算法在运动目标平移、姿态变化、遮挡、光照变化及相似颜色干扰等复杂背景下都能够准确地进行跟踪，对背景干扰具有很强的鲁棒性。

关键词: 加权颜色分布, 不变矩, 自适应融合, 粒子权值, 粒子滤波