CF-DLTrack:A Hybrid Object Tracking Approach Combining Correlation Filter and Deep Learning
Xuexin Liu, Zhuojun Zou, Shuai Tian, Zhifeng Lv, Shun Li, Jie Hao · 2024
Object tracking is a crucial field in computer vision, with significant applications in areas like drone operation and intelligent driving, where real-time performance and high accuracy are essential. Conventional correlation filter trackers offer good real-time performance but have limited accuracy due to inadequate feature extraction. Conversely, deep learning trackers achieve higher accuracy through advanced features but operate at slower speeds. To overcome these issues, we propose a hybrid tracking approach that combines deep learning with correlation filtering. The deep learning tracker supervises the correlation filter tracker, balancing real-time performance and accuracy through a non-equal frame fusion strategy. Extensive testing on OTB100, NFS30, UAV123, and GOT10K datasets confirms that our hybrid tracker significantly enhances speed while maintaining high accuracy, making it ideal for applications that demand quick and precise tracking.