基于压缩感知的长时间实时跟踪技术研究

摘要

Situations such as drift and losing would be easily occurred in compressed sensing-based tracking algorithm under complex conditions like big changes in textures or lightings of the object or transient occlusion.To solve this problem, we propose an improved algorithm which integrates long-term tracking with detection.This algorithm, by introducing cascaded search strategy, can quickly regain the exact location of the object after offset tracking or failure in tracking happens, and effectively reduces the occurrence numbers of the drift and greatly improves the accuracy and stability of tracking.The results got from variant video sequences testing show that the proposed algorithm can accurately detect and track the object in real-time once again in the circumstances such as the object moves fast and irregularly, be partially or completely occluded, or even out of the view of the camera for a moment.%针对基于压缩感知的跟踪算法在目标发生纹理或光照较大变化或短暂遮挡等复杂情况下,容易发生漂移甚至跟丢的情况,提出一种将长时间跟踪与检测相融合的改进算法. 该算法通过引入级联的搜索策略,在目标跟偏或跟丢后,可以快速地重新定位目标的准确位置,有效地减少了漂移发生的次数,很大程度上提高了跟踪的准确性和稳定性. 对不同视频序列的测试结果表明,所提出的算法能够在目标发生快速不规则运动以及目标部分或全部被遮挡甚至在完全离开摄像机视野后,依然能够再次准确地检测并实时跟踪目标物体.