The Determination of Next Best Views
Authors: Connolly · Year: 1985 · Venue: Proc. 1985 IEEE Int. Conf. on Robotics and Automation (ICRA), vol. 2, pp. 432–435 Raw: not obtained — the 1985 scan is closed-access (abstract only via Semantic Scholar, DOI 10.1109/ROBOT.1985.1087372) and the LAN cluster is down (no PDF conversion). Depth: PARTIAL — page built from the S2 abstract plus the authoritative in-corpus description in scott2003view §6.2.2, not from the primary text.
Summary
The origin paper of next-best-view planning. Connolly presents two algorithms that use partial octree models to determine the “best” next view to take in order to obtain a complete model of a scene. Its lasting significance, per the field’s canonical survey, is terminological and conceptual: Connolly first coined the term “next-best-view” (NBV) and framed the problem of choosing the single most informative next sensor pose from a set of candidates on a sphere around the object. We hold this as a historical/foundational anchor for the inspection chapter rather than a method we implement.
Key Claims
From the S2 abstract:
- Two algorithms using partial octree models determine the best next view to take toward a complete scene model.
As described in scott2003view §6.2.2 (secondary, in-corpus):
- Connolly first coined the term “next-best-view” (NBV) — one of the earliest view-planning papers.
- The imaging work space is voxelized and labeled empty, occupied, or unseen, encoded in an octree that also represents the object surface; candidate viewpoints are evenly spaced points on a sphere around the object, all viewing vectors pointing to the origin.
- Planetarium algorithm: a full visibility analysis of the octree for every candidate viewpoint; the area of unseen voxels projected onto the image plane measures the solid angle of unseen space a view would sweep, and the viewpoint with the largest unseen area is chosen as the NBV. It is accurate but has high time complexity (global visibility for all candidates).
- Normal algorithm: simplifies the analysis to the local level by examining octree faces common to both unseen and empty voxels; faster, but handles self-occluding scenes less well.
Method
Not transcribed — primary text not obtained (see Raw). The mechanism, as reported by Scott et al. 2003: a volumetric, generate-and-test view planner — discretize viewpoint space to a sphere, score each candidate by projected unseen area (planetarium) or by unseen/empty octree-face adjacency (normal), and pick the maximizer as the next best view. Do not attribute equations or quantitative results to this page until the primary paper is converted.
Relevance to thesis
Connolly is where our chapter’s central term originates, so it must be cited for provenance of “next-best-view.” Conceptually it is the volumetric, non-model-based end of the view-planning taxonomy (scott2003view): score candidate views by information gain (unseen volume) over a discretized viewpoint set. That is precisely the shape of our mesh-view scorer, with the octree/unseen-voxel gain replaced by a coverage/utility score on the target mesh. It is a supporting historical anchor, not a baseline we run: the 1985 method assumes a small object, a sphere of viewpoints pointing at the origin, and a single-DOF-scale positioning setup, none of which match a free-flying 6-DOF inspector.
Connections
Topics: next_best_view (PAGE NEEDED) · motion_planning · d_optimality · Sources: scott2003view · galceran2013survey
Key Equations / Quotes
“Two algorithms which use partial octree models to determine the ‘best’ next view to take to obtain a complete model of a scene are described.” (Abstract, via Semantic Scholar)
“One of the earliest papers on view planning was by Connolly [1985]. He appears to have first coined the term ‘next-best-view’ (NBV). Connolly presented two algorithms, planetarium and normal…” (Scott, Roth & Rivest 2003, §6.2.2)
Open Questions
- Convert the primary ICRA 1985 paper once the LAN cluster is back — verify the octree labeling, the sphere sampling, and the two algorithms against the source before citing any specifics beyond the term “NBV”.
- Connolly scores by unseen volume/area; our inspection utility is coverage/pose-estimation quality on a mesh — what is the faithful lift of “projected unseen area” to a mesh-view utility for a 6-DOF flying inspector?