基于子空间投影的复杂水下环境运动小目标检测前跟踪方法

针对复杂水下环境运动小目标检测中存在的目标信号强度弱、信杂比低等问题，该文提出基于子空间投影的检测前跟踪(TBD)算法:对原始图像数据截取序列片段，将3维时空片段中的短时运动航迹投影到2维子空间平面；利用2维投影图中平面航迹的形态特征进行初步筛选，提取目标的有效运动区域；将2维平面中的目标短时航迹在局部区域重建3维时序，在3维航迹回溯过程中利用目标运动特征再次筛选目标短时航迹。通过上述分级检测机制，可实现快速高精度的目标短时航迹检测。结合前景检测以及基于层次凝聚聚类(HAC)的长时航迹检测算法，构建了针对运动小目标的完整检测前跟踪方法。最后使用实测声呐图像数据验证了算法的检测精度和检测速度。

Subspace Projection Based Track-Before-Detect Scheme for Small Moving Target in Complex Underwater Environment

The small moving target detection in complex underwater environment is complicated due to the weak target signal strength and low signal-to-clutter ratio. A Track-Before-Detect (TBD) algorithm based on subspace projection is proposed to solve these problems. A sequence motion track fragment is extracted from the original data, and then projected from the 3D space-time onto the 2D subspace. The morphological features in 2D subspace are applied to preliminary screening to remove most of the clutters and locate the local motion areas of the target. The 3D space-time track is reconstructed from 2D subspace in these local motion areas. During the above 3D track backtracking process, the motion continuity characteristics are also extracted to further remove the clutters and select the effective target track fragments. Through the above-mentioned hierarchical processing, a fast and high-precision target track fragment detection algorithm is achieved. By combing this track fragment detection algorithm with the foreground detection and Hierarchical Agglomerative Clustering (HAC) based long-time track detection algorithms, a complete TBD scheme for small moving targets detection is constructed. The accuracy and speed of this detection scheme are verified on the real sonar image data.