旋转摄像机下的快速目标检测算法

The method of rapid moving object detection in rotating camera

对于动态场景下的目标检测,传统的方法已经不能适用。提出一种基于SIFT特征点预测的匹配算法,提高了运动目标检测的速度和准确率。在此过程中,用仿射参数模型结合最小二乘法求解出全局运动参数,再对运动进行补偿。采用ORSA的方法去除外点的影响;采用基于残差图像块的更新策略,实时更新特征点集。最后使用背景差法来实现运动目标的检测并实时的对背景进行更新。该算法不仅保持了SIFT本身的优越性,而且提高了检测速度。实验结果表明该算法可以实时准确的检测出运动目标。

The traditional methods can not be applied for target detection in dynamic scenes. The method is based on the algorithm of the SIFT, the algorithm not only improves the speed but also increases correction ratio. In this process, We using affine parameters, the least square method can precisely compute the motion parameters, that then is leaded to the implementation of motion compensation and the object detection. ORSA algorithm to eliminate the outliers generated in the process of feature points matching, an updating strategy based on residual image is proposed, which updates the feature set sequentially to adjust the change of the background. In the last, background subtraction approach to detect moving objects. The method remains the advantages of SIFT and raises the detection rate significantly. The experiments demonstrate that the method can effectively detect objects while achieving real-time performance.