采用多重约束的无人机导航KLT视觉跟踪算法

Algorithm of vision tracking for UAV navigation based on multi-constraint KLT

特征点跟踪是实现无人机视觉里程计导航的重要技术。针对无人机视频图像帧间运动较大造成帧间的特征点跟踪误差大的问题,提出一种基于时间可逆性约束和双向偏移量约束相结合的多重约束KLT特征点跟踪策略,并在金字塔表示下分层求解跟踪点的偏移量。基于时间可逆性约束,建立新的融合前向跟踪和后向跟踪的目标函数,解算前向跟踪偏移量和后向跟踪偏移量,并构造新的帧间偏移量-双向偏移量,在金字塔分层表示结构下实现偏移量的最优估计。实验结果表明,该方法能够有效地实现帧间特征点的精确跟踪,与同类跟踪算法相比有较好的效果。

Feature point tracking is an important technology to implement the visual odometry for navigation. Aiming at the problem of feature tracking with big errors caused by large motion of the video fixed on UAV, a multi-constraint KLT tracking strategy based on time-reversibility and bi-directional displacement constraint was proposed and the pyramid model was used for the hierarchical displacement computation of the tracking points. The new objective function was set up according to the fusion of forward and backward tracking. A new bi-directional displacement was constructed based on the displacements of forward and backward tracking, and the optimal estimation of the displacements was implemented in the structure of pyramid model. The experiment results demonstrate that the proposed algorithm improves the performance of precise tracking effectively and outperforms the similar tracker.