Abstract: Hybrid particle filter based flame tracking algorithm is proposed in this paper.In order to decrease the computational load of general particle filter,Mean Shift algorithm is embedded into particle filter.At the same time a new adaptive state transition model and an auto-updating strategy of object model are used to improve the performance of hybrid particle filter algorithm.Then,a flame tracking algorithm composed of flame detection and flame tracking is presented.The first phase is flame recognition and the next is flame tracking.If the likelihood of a candidate with the model is smaller than a threshold,it would get into the first phase during tracking.Auto-switch between detection and tracking ensures the validity of flame tracking.The experiment results show that the hybrid particle filter has a good performance.Compared with Mean Shift and general particle filter,it improves the tracking precision.Hybrid particle filter based flame tracking method can track the moving flame under the complex actual environment and get the flame position information.

Key words: particle filter, Mean Shift, flame detection, flame tracking

摘要： 提出一种基于混合粒子滤波的运动火焰跟踪算法。针对通用粒子滤波算法计算量大的问题，提出了混合粒子滤波，将Mean Shift算法嵌入到粒子滤波中，并用自适应运动模型和目标模型自动更新的策略改善算法性能。基于混合粒子滤波提出了火焰识别和火焰跟踪相结合的运动火焰自动跟踪算法，先火焰识别，再火焰跟踪，且跟踪时，如果估计目标与模型的相似度小于阈值则切换到火焰识别阶段。识别与跟踪的相互切换保证了跟踪结果的正确性。实验结果表明混合粒子滤波具有很好的跟踪效果，与粒子滤波和Mean Shift算法相比，提高了跟踪精度；基于混合粒子滤波的火焰跟踪算法能够跟踪复杂环境下的运动火焰，提供火焰的精确位置。

关键词: 粒子滤波, Mean Shift, 火焰识别, 火焰跟踪