Abstract:In recent years, tracking algorithms based on correlation filtering have been widely considered because of their high real-time performance. However, most of these algorithms do not consider the reliability of tracking results; thus, model drift is often a challenge in long-term target tracking. Moreover, the target is significantly occluded or disappears in long-term target tracking, hence the need for research to address such complexities. This paper proposes a long-term tracking framework comprising a tracking and re-detection module. The tracking module is based on the efficient convolution operator target-tracking algorithm (ECO) (Danelljan et al. in Proc IEEE Conf Comput Vis Pattern Recognit, 1). To redetect a lost target, a tracking uncertainty estimation method is developed that evaluates the tracking results of each image. Furthermore, an adaptive model-updating method is proposed, which can reduce the number of model updates and improve the robustness of the tracking algorithm. The model is inspired by long and short-term memory pool mechanisms of the brain, applying both mechanisms to the traditional tracking algorithm to improve long-term tracking. The memory model is effectively integrated into a brain-inspired visual long-term tracker through mutual learning and inspiration from computer and biological vision. In addition, the brain-inspired visual model can be made bio-cognitive on a small hardware platform with limited computing power. Central processing unit (CPU)-based experiments using two data sets, UAV20L (Mueller et al. in Eur Conf Comput Vision, 2) and UAV123 (Wu et al. in Proc IEEE Conf Comput Vis Pattern Recognit, 3), confirmed that the proposed method runs faster than 30fps. Furthermore, the long-term tracking test using the UAV20L data set showed that the proposed method performs better than any other method by 39%. Compared with conventional tracking methods, the proposed method has better performance in terms of coverage rate and positioning accuracy.

Long-term Tracking Based on Deep Learning

Fast-Tracker 2.0: Improving Autonomy of Aerial Tracking with Active Vision and Human Location Regression.

Track Without Appearance: Learn Box and Tracklet Embedding with Local and Global Motion Patterns for Vehicle Tracking

A Compensatory Algorithm for High-Speed Visual Object Tracking Based on Markov Chain

Long-term tracking algorithm using deep features and a single shot multibox detector.

A Top-View Multiple People Tracking System Based on Newest YOLOv5 and DeepSort Using Depth Data

Re-detection and distractor association from a global perspective: A long-term tracking system

Long-term object tracking based on joint tracking and detection strategy with Siamese network

Fast and Sub-Pixel Precision Target Tracking Algorithm for Intelligent Dual-Resolution Camera.

Effective Local and Global Search for Fast Long-term Tracking

Target-Aware Deep Tracking

A Perceptually Adaptive Long-Term Tracking Method for the Complete Occlusion and Disappearance of a Target

Long-Term Tracking Algorithm with the Combination of Multi-feature Fusion and YOLO.

Unsupervised Deep Representation Learning for Real-Time Tracking

Video Object Tracking Based on SSD and Camshift

Deep visual tracking: Review and experimental comparison

Detection and tracking of human track and field motion targets based on deep learning

Robust Visual Tracking Method via Deep Learning

Real-time tracking based on deep feature fusion

Visual Object Tracking with Adaptive Template Update and Global Search Augmentation

Real-Time Multiple Object Tracking with Discriminative Features