基于SIFT特征与预测的运动目标检测算法

Detection algorithm of moving object based on predicted SIFT feature

为了在动态场景下对运动目标进行快速检测,提出一个改进的SIFT特征匹配的检测算法。首先采用SIFT方法提取匹配的特征点;然后为全局运动建立起旋转参数模型,并使用RANSAC方法排除外点的影响,运用最小二乘法求解全局运动参数;最后利用基于残差图像块的更新策略对特征点进行更新。该算法是基于预测的SIFT特征点匹配算法,不仅保持了SIFT方法的优越性能,而且提高了检测目标的速度。与块匹配算法的实验结果对比表明,该算法可以实时准确地检测出运动目标。

An improved detection algorithm of SIFT feature marching is provided for rapid detection to moving object in dy？namic scene. The matched feature points are extracted by SIFT method,and the rotation parameter model is established for global motion. The influence of exterior points is eliminated by RANSAC method. The global motion parameters are solved by the least square method,and the feature points are updated by the updating strategy based on residual image block. This matching algo？rithm is based on the predicted SIFT feature points,which can remain the high performance of SIFT method and improve the de？tection rate of the object. Compared with the experimental results of block matching algorithm,it demonstrates that this algo？rithm can detect moving object accurately and in real？time.