无人机对地运动目标跟踪算法

Method for Ground Moving Target Tracking of UAV

摘要:

提出一种基于改进相关滤波器的无人机对地运动目标跟踪算法。该算法提取目标梯度方向直方图特征、颜色特征、深度特征等多种特征,利用空间可靠性图改进相关滤波器,提高对不规则形状目标的跟踪性能,基于高效卷积算子进行特征降维和紧凑样本空间构建,降低数据冗余。根据相关滤波器得到的目标特征响应确定目标跟踪框的位置,实现目标跟踪。经实验验证:该算法的跟踪平均中心距离误差为16.07,跟踪成功率曲线下面积参数为0.54,均优于目前其他先进算法;跟踪平均帧率最高可达44.09帧/秒,具有较好的实时性。

Abstract:

An algorithm of tracking moving target on the ground by unmanned aeria.vehicle (UAV) based on improved correlation filter is proposed. Multiple features of thetarget, such as histogram of gradient ( HOG ) feature, color names ( CN) feature.convolutional neural network ( CNN) feature are extracted. To improve the trackingperformance of irregular objects, the spatial reliability graph is used to improve thecorrelation filter. Based on efficient convolution operator, feature dimension reduction andcompact sample space are constructed to reduce data redundancy. The position of the targettracking box is determined according to the target characteristic response obtained by thecorrelation filter. So the target tracking is accomplished. Experimental results show that themean centre location error (CLE) is 16.07, the area under curve(AUC) of tracking successrate is 0. 54. Both evaluation parameters are superior to other advanced algorithms. Thealgorithm can track small moving target accurately in real time, with an average frame rateof 44.09 frames per second.