基于差分和特征不变量的运动目标检测与跟踪

New method for detecting and tracking of moving target based on difference and invariant

提出了一种基于改进的图像差分算法与特征不变量匹配的目标识别方法。通过三帧差值法获得了更完整清晰的目标轮廓,并基于该轮廓信息构造了一个具有平移、大小和旋转不变性的特征不变量;然后提出动态极值匹配法,利用特征曲线的极值信息点进行识别匹配,并动态替换原特征模版。实验结果表明,该方法能够准确识别目标,显著地提高识别跟踪效率,并且适用于检测运动姿态发生变化的目标。对于分辨率为288×352像素,每像素8位量化的序列图像,处理每帧图像平均用时0.011 74 s,其中特征提取与匹配过程平均用时0.005 476 s,能够实现对运动目标的实时分析,可同时满足运动目标识别跟踪中实时性和准确率的要求。

A new method for detecting and tracking moving target is proposed based on combination of image difference and feature invariant. A more complete and clear contour is acquired using three-frame difference. And a feature invariant is constituted without influence of object＇s size change, shifting, and rotation based the contour. Then a dynamic extremum-matching method is proposed for recognizing and matching the object based on extreme point information and replacing the former mold-plate with the new one. The experimental results indicate that the method can successfully detect moving target and accurately estimate its trajectory in the image sequence with less memory even when the object figure is changed during movements. Moreover, it is efficient and adaptable for real-time target detecting and tracking, and it can meet the demands in real-time detecting and tracking moving targets.