基于可见光与红外图像特征融合的目标跟踪

摘要

Aiming at the problem that the accuracy of tracking object is not high when with a single image source, CAMShift(Continuously Adaptive Mean Shift) tracking algorithm is improved by using different characteristics of the tracked target in the visible images and infrared images. Firstly,“color-gradient back projection”is selected as the improved target model in visible image, and“gray-gradient back projection”is selected as the improved target model in infrared image. Then the coefficient of qi which is got by using the improved CAM Shift tracking algorithm in visible images and infrared images respectively is used to judge the effect of the two images tracking. The weights of two images are adjusted adaptively by the coefficient of qi. Finally, the feature fusion image and the location of object are got according to the respective weight. The experimental results show that, for visible and infrared image of 320pixel×240pixel, the object tracking algorithm which is based on feature fusion by visible images and infrared images can get much accurate location of tracking target under complex background in which the accuracy of tracking object is 0.5pixel, and the velocity of tracking object is 30-32 ms/frame.%针对单一图像源下目标跟踪精度不高的问题，利用跟踪状态下的目标存在于可见光与红外图像中的特征对连续自适应均值移动跟踪算法做出改进。首先选取可见光图像的“颜色梯度背投影”作为改进的目标模型，选取红外图像的“灰度梯度背投影”作为改进的目标模型；然后根据可见光序列图像和红外序列图像各自进行连续自适应均值移动跟踪算法得到的对应的qi系数判定两种图像跟踪的效果，对两种图像的权重进行自适应调整，得到这两种图像的特征级融合图像和跟踪结果。实验结果表明，对于320像素×240像素的可见光和红外图像，基于可见光与红外图像特征融合的目标跟踪算法在复杂背景下能够较准确的跟踪目标，目标跟踪精度为0.5像素，跟踪速度为30~32 ms/帧。