Ellipsoid SLAM with Novel Object Initialization

Object SLAM With Robust Quadric Initialization and Mapping for Dynamic Outdoors

“…On the other hand, the generic model methods can be divided into parametric and nonparametric approaches. Parametric approaches represent objects using regular 3D forms such as cubic boxes [16], [17] and dual quadrics [5], [6], [7], [8], [9], [10]. These methods tightly constrain the parameters of the 3D model by using the 2D bounding box corresponding to the object.…”

“…Dual quadrics, in particular, have gained attention due to their compact mathematical parameterization and ease of manipulation within the framework of projective geometry [5]. Recent research has focused on using dual quadrics as object representations to improve system robustness and localization accuracy [5], [6], [7], [8], [9], [10]. In contrast, nonparametric approaches reconstruct and represent objects using a combination of geometric structures such as surfels [15], [27], [33], voluments [34] and clusters [28], [35].…”

“…within the environment [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20]. This semantic information promotes target-oriented tasks like obstacle avoidance [13], camera localization [10], [11], and robust relocalization [9], [12].…”

“…Accurate object representation is a key issue in objectoriented SLAM research, commonly used object representation can be divided into the prior object model [18], [19], [20], and the generic object model [5], [6], [7], [8], [14], [15], [16], [17]. The prior object model methods rely on the prebuilt CAD model databases [18], [19] or point cloud models [20].…”

“…In contrast to the cubic box, the ellipsoid can be compactly parameterized and easily manipulated within the framework of projective geometry, which has attracted attention in recent work due to the closed surface of the ellipsoid being meaningful for object landmarks [5], [6], [7], [8]. However, the accuracy and robustness of current quadric-based SLAM algorithms are not ideal, especially the quadric initialization process, which is limited by the parameter coupling of the direct linear solution method [5], the necessity of point cloud fitting [14], or the commonly used decoupled orientation methods rely on planar assumptions [6], [8], which limits their applications in realworld scenarios. Therefore, we propose a novel robust and accurate quadric initialization algorithm for outdoor scenes.…”