More about HKUST
A Survey on Rigid Object Pose Estimation
PhD Qualifying Examination Title: "A Survey on Rigid Object Pose Estimation" by Miss Na FAN Abstract: Rigid object pose estimation is a fundamental task in computer vision with applications across various domains such as robotics, augmented reality (AR), and autonomous driving. Given single or combinations of input types from RGB images, depth images, point clouds, and text prompts, along with an optional CAD model of the target object, pose estimation aims to determine the relative transformation (rotation and translation) of the object's coordinate frame with respect to the camera's coordinate frame and potentially estimate the object size. The field has rapidly evolved from traditional methods relying on handcrafted features to learning-based approaches, progressing from instance-level to category-level pose estimation and further to unseen pose estimation. More recently, with pre-trained large vision and language models, it has advanced to handle more challenging open-vocabulary object pose estimations. Persistent challenges include the reliance on annotated data, robustness in complex scenarios (e.g., cluttered environments, occlusions), handling objects with special properties (e.g., symmetry, transparency), inference efficiency, and generalization to new objects. This survey begins by introducing the problem definition of rigid object pose estimation and the evaluation metrics. Subsequently, we categorize existing methods into instance-level, category-level, and unseen object pose estimations, introducing the related datasets and methods at each level. We categorize the methods by their approaches and input types and introduce in detail the representative methods, analyzing the technical details, strengths, and weaknesses. In conclusion, we summarize these methods and discuss potential future directions in the field of rigid object pose estimation. Date: Friday, 6 December 2024 Time: 1:00pm - 3:00pm Venue: Room 3494 Lifts 25/26 Committee Members: Dr. Qifeng Chen (Supervisor) Prof. Qiong Luo (Chairperson) Dr. Brian Mak Dr. Dan Xu