Abstract:When the target is partially occluded or undergoes drastic appearance change, drift is prone to occur in traditional correlation filter-based trackers and ultimately leads to tracking failure. This paper proposes a robust tracking algorithm, which can be applied to practical engineering. The proposed algorithm divides all training samples into different training sets based on their similarities. An independent filter is trained for every training set. In each frame of the tracking process, the similarity between the candidate region and any training set's training samples is calculated to select the most matched filter for target location. As a newly introduced training sample, a feature map within the current frame's estimated target bounding box is assigned to the most similar training set to update the most matched filter. The updating of each filter is relatively independent. Each filter corresponds to one kind of target's typical appearance, and thus, the proposed tracking algorithm can memorize a variety of typical target appearances that appeared in the past and adapt to target's discontinuous appearance change. The detector improves tracking accuracy and assists occlusion judgment. After occlusion, the target may reappear with a different posture. As long as similar postures have been captured in the previous tracking process, the target can be retrieved accurately while not being confused with other objects of the target's category. We evaluated our method on massive videos shot in actual engineering sites, and the result demonstrates that the proposed tracking algorithm can handle occlusion and target's posture changes very well and significantly outperforms other tracking algorithms.

TLD Target Tracking Algorithm Based on Dynamic Capture

An Improved TLD Visual Target Tracking Method

Fast object tracking method based on TLD algorithm

TLD Target Tracking Fused with GMS Detection and Confidence Discrimination

A Visual Object Tracking Algorithm Based on Improved TLD

Research on Target Tracking Based on TLD Algorithm

PE-TLD: Parallel Extended Tracking-Learning-Detection for Multi-target Tracking.

An Improved TLD Tracking Method Using Compressive Sensing

An Adaptive Tracking Algorithm Based on Mean Shift

A Target Tracking Algorithm Based On Mean Shift With Feature Fusion

Object Tracking Algorithm with Sparse Prototype Based on TLD

Target tracker with masked discriminative correlation filter

Htld: A Human-in-the-loop Target Detection and Tracking Method for UAV

An extended KCF tracking algorithm based on TLD structure in low frame rate videos

Real-time long-term tracker with tracking–verification–detection–refinement

Real-Time Interpretation Method For Shooting-Range Image Based On Position Prediction

Discriminative Correlation Filter for Long-Time Tracking

Target tracking based on improved accelerated gradient under tracking-learning-detection framework

Pedestrian detection and tracking based on TLD framework

Real-time Target Object Tracking Via Dual-Directional Scaling State Predicting

A Detection Aided Multi-Filter Target Tracking Algorithm.