自适应结构化稀疏表示的海上目标跟踪研究

ON MARITIME TARGETS TRACKING BASED ON ADAPTIVE STRUCTURED SPARSE REPRESENTATION

针对海上目标图像背景结构复杂,目标间相似度大,且遮挡和碰撞时常发生的特点,提出一种基于结构化局部稀疏表示的对角块外观模型构造方法,通过对外观表现稳定的图像块赋予较大权重来增加跟踪的稳定性。建立递推的目标跟踪模型,通过卡尔曼滤波算法解决每帧瞬时目标和跟踪结果之间的后验稀疏向量表示问题。最后实现帧目标搜索,瞬时目标与基函数词典之间残差最小的目标作为最优跟踪结果,同时采用增量主成分分析法重构跟踪结果以减少遮挡影响。将此方法与经典的目标跟踪算法进行比较,实验表明此方法具有较好鲁棒性和稳定性。

For the characteristics of maritime target image including complex background structure, great similarity between targets, and frequently happened occlusions and collisions, we propose a construction method for diagonal block appearance model which is based on adaptive structured sparse representation. First, it increases the stability of tracking by assigning greater weight to image block with stable appearance performance. Then it builds a recursive target tracking model, through Kalman filter algorithm it solves the posteriori sparse vectors representation of each frame between instantaneous goal and tracking result. Finally, it realises the frame target search. The target with minimal residual between instantaneous target and basis function dictionary is taking as the optimal tracking result, and meanwhile the incremental principal component analysis is used to reconstruct tracking result for reducing the influence of occlusion. Comparing this method with the classical target tracking algorithms, experiment showed that our algorithm has better robustness and stability.