Abstract

Target tracking has always been an important research direction in the field of computer vision. The target tracking method based on correlation filtering has become a research hotspot in the field of target tracking due to its efficiency and robustness. In recent years, a series of new developments have been made in this research. However, traditional correlation filtering algorithms cannot achieve real-time tracking in complex scenes such as illumination changes, target occlusion, motion deformation, and motion blur due to their single characteristics and insufficient background information. Therefore, a scale-adaptive anti-occlusion correlation filtering tracking algorithm is proposed. First, solve the single feature problem of traditional correlation filters through feature fusion. Secondly, the scale pyramid is introduced to solve the problem of tracking failure caused by scale changes. In this paper, two independent filters are trained, namely the position filter and the scale filter, to locate and scale the target, respectively. Finally, an occlusion judgment strategy is proposed to improve the robustness of the algorithm in view of the tracking drift problem caused by the occlusion of the target. In addition, the problem of insufficient background information in traditional correlation filtering algorithms is improved by adding context-aware background information. The experimental results show that the improved algorithm has a significant improvement in success rate and accuracy compared when with the traditional kernel correlation filter tracking algorithm. When the target has large scale changes or there is occlusion, the improved algorithm can still keep stable tracking.

Highlights

  • With the continuous development and progress of modern technology, the discipline of artificial intelligence has entered a new era, and the use of artificial intelligence technology to replace traditional manual labor has become a development trend

  • Our improved scale-adaptive anti-occlusion correlation filter algorithm SACF and scale-adaptive anti-occlusion correlation filter for context-aware SACF_CA are compared with several current classical correlation filter target tracking algorithms, including several algorithms of KCF, DSST, CSK, CT, SAMF, TLD, and the comparison chart is as Figure 3: Under the aforementioned hardware and software experimental conditions, the average FPS of our proposed improved algorithm SACF is 70.138

  • The tracking accuracy and success rate are improved by 6.2% and 5.9%, respectively, compared with the original KCF algorithm, the tracking accuracy reaches 79.9% and the success rate reaches 67%

Read more

Summary

Introduction

With the continuous development and progress of modern technology, the discipline of artificial intelligence has entered a new era, and the use of artificial intelligence technology to replace traditional manual labor has become a development trend. In some image sequences the target moved out of the field of view, in the actual application scenario there may be complex climate conditions change, excessive external noise and other factors of interference. These are the main technical challenges of the current tracking algorithms to ensure tracking accuracy while taking into account the real-time tracking performance;. (3) Search box size selection problem: If the search box is too small, it is not easy to detect fast-moving targets; if the search box is too large, it will introduce a lot of useless background information, and even some backgrounds similar to the target will interfere with the target tracking process, leading to the degradation of the model and the phenomenon of tracking drift. (4) For some complex application scenarios, the discriminative power is enhanced by adding contextual information blocks

Related Work
Model Updates
Feature Extraction Based on Multi-Feature Fusion
Scale Adaptive Evaluation
Model Update Strategy
CCoonntteexxtt--AAwwaarreeCCoorrrreellaattiioonnFFiilltteerrss
Evaluation Indicators
Quantitative Analysis
MB SV OV LR BC OCC IPR OPR DEF FM
Qualitative Analysis
Conclusions
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call