Click to read more
Overview
Introducing BOBBY2, a novel high-speed single object tracker designed to be robust against non-semantic distractor exemplars. BOBBY2 features a groundbreaking exemplar buffer module that sparsely caches the target's appearance over time, allowing for adaptation to potential target deformation.
Key Features
Exemplar Buffer Module: Caches target appearance across time, enabling redundancy and adaptation to target deformation.
Efficient Training: Trained using an augmented ImageNet-VID dataset with the one cycle policy, reaching convergence in less than 2 epochs.
Robust Performance: Maintains near-optimal accuracy even when the buffer is filled with distractors.
High-Speed Operation: Utilizes a stripped-down AlexNet with 63% fewer parameters, achieving 85 FPS.
Generalizability: Achieves competitive results on the GOT-10k dataset and the custom challenging TU-3 UAV dataset without fine-tuning.
Project Members (2018-2020)
Lee Kiefer
Publications
K. Lee, S. K. Phang and W. J. Chew, "Development of low computational power vision tracking algorithm on embedded system," AIP Conference Proceedings, vol. 2137, no. 1, p.030004, 2019.
K. Lee, J. J. Tai and S. K. Phang, "Bobby2: Buffer based robust high-speed object tracking," arXiv preprint arXiv:1910.08263, 2019.
Click to read more
Overview
The Closest Obstacle Avoidance and A* Algorithm (COAA*) is an innovative approach to obstacle avoidance and navigation (OAN) for multirotor Unmanned Aerial Vehicles (UAVs). COAA* seamlessly integrates the speed of offline algorithms with the flexibility of online methods, eliminating the need for prior map knowledge.
Key Features
Computational Efficiency: Designed to operate on onboard companion computers, COAA* is computationally lightweight.
Global Minimum Convergence: Guarantees convergence to a global minimum for navigational trajectories.
Versatility: Easy calibration and integration for various mobile robots beyond UAVs.
Performance-Aware: Takes UAV performance limits into account for optimized operation.
Project Members (2018-2019)
Jun Jet Tai
Publications
J. J. Tai, S. K. Phang and F. Y. M. Wong, "COAA* - An Optimized Obstacle Avoidance and Navigational Algorithm for UAVs Operating in Partially Observable 2D Environments," Unmanned Systems, vol. 10, no. 2, pp. 159-174, 2022.
J. J. Tai, S. K. Phang and Y. M. F. Wong, "Optimized autonomous UAV design with obstacle avoidance capability," AIP Conference Proceedings, vol. 2233, no. 1, p.020026, 2020.