基于位置修正机制和模型更新策略的跟踪算法

为了实现目标跟踪算法在无人机应用中的高实时性和高成功率,针对目标遮挡和目标形变量大导致的跟踪丢失问题,提出了一种基于相关滤波的改进算法,算法包含自适应位置修正机制和模型更新策略.该算法提取目标区域的方向梯度直方图特征(HOG)训练滤波器,预测下一帧的目标位置.当预测位置不满足高置信度条件时,融合颜色命名特征(CN)对位置进行修正.为了提高算法的运行效率,对融合后的特征进行主成分分析(PCA)降维处理.利用平均峰值相关能量、多峰检测及最大响应值进行模型更新.实验中将改进算法与近年来的优秀算法进行对比.结果表明,所提算法在目标被遮挡和形变量大的场景中,跟踪精度更高.

Tracking Algorithm Based on Location Correction Mechanism and Model Update Strategy

To achieve high real-time performance and high success rate of the target tracking algorithm in UAV applications,we propose an improved algorithm based on correlation filtering for tracking loss caused by target occlusion and target large deformation.The algorithm includes an adaptive position correction mechanism and a model update strategy.By extracting the histogram of the oriented gradient features of the target region,we train the filter and predict the target position of the next frame.When the position does not satisfy the high confidence condition,the fusion color naming features correct the position.To improve the efficiency of the algorithm,principal component analysis dimension reduction processing is performed on the merged features.The filter model is updated using average peak correlation energy,multi-peak detection,and maximum response value.In the experiment,we compare the improved algorithm with excellent algorithms that have been developed in recent years.Results show that the proposed algorithm has higher tracking accuracy in the scene where the target is occluded and target’s large deformation.