Radar Single Extended Object Tracking Based on Correlation Filter

Abstract

In traditional radar object tracking algorithms, targets are regarded as points in tracking filtering algorithms. With higher resolution in range and azimuth directions, a target can occupy multiple resolution units, so radars can obtain high-resolution observation results. When point targets turn into extended objects, the traditional point target tracking filtering algorithms can hardly achieve stable tracking to the extended object due to the increasing complexity in data association. In this paper, a radar single extended object tracking algorithm is proposed based on correlation filtering algorithm in the computer vision field. This algorithm normalizes radar echoes to get the visual radar range-azimuth image firstly. Then, this image is transformed into the Cartesian coordinate system from the polar coordinate system. At last, the target appearance of the extended object are online learned by the correlation filter to realize the target tracking based on filtering response diagram. Real data based experimental results demonstrate that this algorithm can obtain the object location and size in each frame. Compared with traditional filtering algorithms, this algorithm is more accurate and robust.

Keywords

• |time-sensitive target|correlation filtering algorithm|filter response|extended object|online learning|target tracking|radar