基于Camshift自适应多特征模板的视频目标跟踪

Adaptive multi-feature template video target tracking based on Camshift algorithm

Camshift算法实时性高,计算量小,在目标跟踪领域应用效果良好。但其仅依靠颜色模型的特点使得在噪声大、照度不均的井下视频目标跟踪中易造成目标丢失。通过在Camshift基础上建立多特征融合的模板自适应更新算法,实现边缘、纹理等特征的融合,制定特征贡献度规则,在环境变化时根据不同特征贡献度的不同自适应分配权重,更新模板。实验结果表明:新算法抗干扰能力强,特征间互补不足,跟踪准确,在煤矿复杂环境井下视频目标跟踪中有良好应用前景。

Camshift algorithm has high realtime performance and low computation, so it is widely used in target track ing field. Under the noisy and uneven illumination coal mine environment, Camshift algorithm will lose target easily, because it relies on color model only. A new adaptive template update model proposed based on the Camshift algorithm and multifeature, such as edge, texture and other features. While environment altering, features can change weight ra tionally by their different contributions, and the template update adaptively. Experiment showed that new algorithm has racked accuracy, high antiinterference ability and complement between different features. It has a good prospect in ob ject tracking under coal mine complex environment.