PaperlayerView exact revision on arXiv

Source-linked AI summary

Immersive Focus+Context Techniques to Assist in Interpretation of High Density Forest Point Clouds

Spencer O'Keeffe, Jan Rombouts, Bruce H Thomas, Michelle Balasso, Jim O'Hehir, Andrew Cunningham

arXiv:2610.22878v1cs.HC

TL;DR

Dense forest point clouds are difficult to assess, while forestry tasks such as tree selection and stand assessment call for greater precision. This paper develops three VR Focus+Context techniques and evaluates them in two user studies. VR outperformed desktop in the first study, and the techniques outperformed the control in the second.

  • Problem

    Dense forest point clouds pose scale and legibility challenges, while traditional tree selection and stand assessment rely on field judgements or extrapolation from small samples.

    Source: §1 · p. 1, §2.2 · p. 2

  • Method

    The authors developed three VR Focus+Context techniques and evaluated VR against desktop, then the techniques against a control in forestry-task analogues.

    Source: §1 · p. 2, §5.0.1 · p. 6

  • Results

    VR outperformed desktop across three performance metrics, and the Focus+Context techniques outperformed the control in every performance metric.

    Source: §3.2 · p. 4, §5.3 · p. 8

  • Takeaways & Limitations

    The authors conclude that Focus+Context techniques are viable for forestry analytics and may apply to other fields with similar spatial data.

    Source: §6 · p. 8

    The study used lower-resolution data for the control and should be expanded to other task analogues, such as tree species and age classes.

    Source: §5.3 · p. 8

Abstract

from arXiv · show

Virtual Reality (VR) is changing how we interact with complex spatial datasets. VR has the potential within forestry to help analyse remotely sensed spatial data for inventory management. However, challenges of scale and legibility of dense forest point clouds must be addressed. To do so, we present three VR Focus+Context (F+C) techniques: Wedgelight, Remote Lantern, and Ray Lantern. These techniques present point cloud data using a constrained highdetail focus within a broader abstracted context representation, manipulated using egocentric and exocentric interactions. We conducted two user studies. The first informed the design of the F+C techniques. The results from the first study indicate that VR significantly outperforms desktop displays in basic analysis of forestry point clouds, with improvements to both time and accuracy. The second study evaluated the performance of the resulting F+C techniques in a task analogous to real-world forestry operations. Our F+C techniques significantly outperformed current approaches, confirming their viability for facilitating detailed and effective forest assessments. These findings indicate that VR has significant advantages for spatial data analysis tasks in forestry, making a strong argument for its adoption in future analytical frameworks. The F+C visualisation techniques we introduced show considerable promise for analysis of high density point clouds, opening new avenues for development in immersive analytics. Supplemental materials (code, scans, results) are available at https://osf.io/hqc9u.

1 INTRODUCTION

Dense forest point clouds could support more precise forest assessment, but their scale and legibility make effective VR interaction challenging. The paper introduces three VR Focus+Context techniques and evaluates VR against desktop displays and the techniques against a control.

1 INTRODUCTION

  • Traditional tree selection and stand assessment rely on field judgements or measurements of a small subset of trees followed by extrapolation.The paper identifies both processes as benefiting from higher precision.Source: §1 · p. 1
  • The paper introduces Wedgelight, Remote Lantern, and Ray Lantern, combining a detailed, user-controlled focus with an abstracted forest context.The focus uses high-density LiDAR scans, while the context uses AI-derived tree features; interactions use egocentric or exocentric controls.Source: §1 · p. 2
  • The formative study compares VR with desktop displays, and a subsequent study evaluates the techniques in a task analogous to tree thinning.The authors assess performance, movement, and participant feedback in the later study.Source: §1 · p. 2
  • The paper’s contributions include empirical comparisons, the three F+C techniques, shared data and scripts, and an evaluation of the techniques in a forestry-operation analogue.The authors state that the techniques may also apply to other domains involving dense point clouds.Source: §1 · p. 2

2 BACKGROUND

Prior work describes benefits of immersive displays for complex spatial analysis while identifying persistent challenges in dense forestry point clouds. Focus+Context offers a way to inspect local detail while retaining surrounding spatial information.

2.1 Immersive Analytics

  • Immersive analytics studies how human-computer interaction and immersive displays can support complex data analysis, including applications beyond forestry.Prior examples include materials and biological datasets, as well as forestry scan measurement and annotation.Source: §2.1 · p. 2

2.2 VR in a Forestry Data Context

  • Forestry point clouds can span hectares, with scattered trees and overlapping branches that obscure structures needed for assessment.Prior work also identifies point density, segmentation, and missing color as factors affecting assessment, while scale transitions and level-of-detail design remain challenges.Source: §2.2 · p. 2
  • Rendering high-resolution point clouds in VR may require preprocessing or display techniques, but reducing detail can blur distinctions or sacrifice information.Prior approaches include voxelization, level-of-detail systems, area selection, culling, and point-cloud compression.Source: §2.2 · p. 2

2.3 Focus+Context

  • Focus+Context lets users inspect part of a large dataset without losing track of its relation to the whole.Related work describes immersive F+C as supporting focus on important information while retaining contextual information in the periphery.Source: §2.3 · pp. 2–3

3 STUDY 1: POINT CLOUD COMPREHENSION ON DESKTOP VERSUS VR

The formative study compares desktop and VR displays on forestry point-cloud comprehension tasks, with display modality and question type as its independent variables. It measures performance, interaction, and feedback under controlled display and input conditions.

3 STUDY 1: POINT CLOUD COMPREHENSION ON DESKTOP VERSUS VR

  • The study compares desktop and VR on forestry comprehension questions designed as analogues of real-world tasks, using expert-informed questions and real scans.Participants completed six question types with four variations each, for 24 tasks.Source: §3.0.1 · p. 3, §3 · p. 3

3.0.1 Task

  • Tasks covered counting, branching, tallest-tree and tallest-cluster identification, split stems, and matching identical trees.Participants could rotate, scale, and translate the point cloud before selecting an answer.Source: §3.0.1 · p. 3

3.0.2 Independent Variables

  • The two independent variables were display condition and question type, and they were counterbalanced to reduce fatigue or learning effects.The conditions were desktop and VR; question types were counting, branching, tallest, tallest cluster, split, and mirror.Source: §3.0.2 · p. 3

3.0.3 Dependent Variables

  • Dependent measures included correctness, counting error, observation time, frame-of-reference transformations, mental effort, usability, and feedback.Transformation measures included translation, rotation, scale, and resets recorded every 0.1 seconds.Source: §3.0.3 · p. 3

3.0.4 Hypothesis

  • The hypotheses predicted that VR would improve performance and be preferred, with its impact varying by task type.The study also tested whether participants preferred VR over desktop.Source: §3.0.4 · p. 3

3.0.5 Apparatus

  • Both conditions used the same laptop and gamepad, while the desktop display was down-sampled to match the headset’s effective resolution.Participants sat for desktop tasks and could sit or stand in VR, remaining consistent within that condition.Source: §3.0.5 · p. 3

3.0.6 Procedure

  • Participants trained on desktop and then VR, completed every task in each condition, and were asked to work as quickly and accurately as possible.There was no task time limit; participants completed post-condition surveys and a post-study questionnaire.Source: §3.0.6 · p. 3

3.0.7 Participants

  • The study recruited 24 university staff and students aged 19–37, including 16 men and 8 women.Eighteen participants had a computer-science background; the remainder came from several other fields.Source: §3.0.7 · p. 3

3.1 Study 1 Results

VR performed better than desktop on correctness, counting accuracy, and cloud observation time, with no detected interaction between display condition and question type. Participants also reported lower effort and higher usability in VR, and generally preferred it.

3.1 Study 1 Results

  • VR outperformed desktop across the study’s three performance metrics, with faster, more accurate responses and better counting accuracy across task types.The authors attribute the depth-perception advantage to stereoscopic view and structure-from-motion; they found no interaction between display and question type.Source: §3.1 · p. 4

3.1.1 Correctness and Tree Counting Accuracy

  • Correctness was higher in VR (M = 0.70, SD = 0.46) than desktop (M = 0.64, SD = 0.48), and VR counting relative error was lower (M = 3.58, SD = 4.96) than desktop (M = 6.36, SD = 9.25).Lower values indicate greater tree-counting accuracy.Source: §3.1.1 · p. 4, Figure 2 · p. 4

3.1.2 Cloud Observation Time

  • Cloud Observation Time was lower in VR (M = 32.1, SD = 30.9) than desktop (M = 37.1, SD = 44.5).The reported display-condition difference was significant (t(1139) = 2.582, estimate = 4.97, p = .010).Source: §3.1.2 · p. 4, Figure 2 · p. 4

3.1.3 Frame-of-Reference Transformation

  • Desktop users made more cumulative frame-of-reference translations and rotations than VR users.Translation was 2.25 m for desktop and 0.99 m for VR; rotation was 0.964 for desktop and 0.4116 for VR.Source: §3.1.3 · p. 4, Figure 3 · p. 4

3.1.4 Participant Feedback

  • Participants reported lower mental effort in VR (M = 3.29, SD = 1.24) than desktop (M = 4.32, SD = 1.22), and higher usability scores in VR (83.6) than desktop (71.8).Twenty-three of 24 participants preferred VR for enjoyment, point-cloud viewing, and overall preference; 20 of 24 preferred it for perceived accuracy.Source: §3.1.4 · p. 4, Figure 3 · p. 4

3.2 Discussion of Study 1 Results

VR outperformed desktop displays across the study’s performance metrics, with no interaction effect across question types. Participants also preferred VR, while the formative study had limitations in participant balance, interaction, and display-resolution comparison.

3.2 Discussion of Study 1 Results

  • VR outperformed Desktop across all three performance metrics, with no interaction effect across question types.The authors attribute the depth-perception advantage to stereoscopic view and structure-from-motion compared with manual desktop manipulation.Source: §3.2 · p. 4
  • VR was the majority preference across feedback categories and had higher usability and lower mental effort than Desktop.Participants also reported VR was easier and more enjoyable for the assigned tasks.Source: §3.2 · p. 4
  • The formative study had an unbalanced participant gender ratio, controller input, and no comparison with higher-resolution desktop displays.The headset-matched resolution precluded testing higher-resolution desktop displays; one participant suggested hand interactions would feel more natural in VR.Source: §3.2 · p. 4

4 FOCUS+CONTEXT TECHNIQUES FOR POINT CLOUD FOREST ASSESSMENT

The section presents Focus+Context visualisation techniques shaped by study observations and forestry-expert feedback. They combine an abstracted forest overview with a high-resolution focus, implemented to support forestry assessment in VR.

4.1 Requirements

  • Four forestry specialists informed requirements for large datasets, tree-feature identification, performance, annotation, assessment tracking, responsive visibility control, and frame-of-reference transformation.The requirements drew on informal discussions with inventory and resource-management specialists and feedback from the formative study.Source: §4.1 · pp. 4–5

4.2 Visualisation Design

  • Focus+Context aligns a high-resolution dataset within a simplified context, preserving geospatial context while displaying detail only inside a defined focus.The approach is motivated by the importance of nearby-tree conditions alongside the tree under assessment.Source: §4.2 · p. 5

4.2.1 Frame-of-Reference

  • The frame-of-reference defines the virtual position and projection of focus and context, and users can independently adjust its scale and rotation.The controller joystick manipulates the frame-of-reference independently of the user’s position.Source: §4.2.1 · p. 5

4.2.2 Context

  • Low-resolution fitted cylinders represent estimated stem points at fixed heights, giving an overview of tree shape, size, and location with minimal performance overhead.The context was generated from a forest scan processed with FSCT, which the authors consider sufficient for geospatial continuity despite ongoing AI-model development.Source: §4.2.2 · p. 5

4.2.3 Focus

  • The focus displays full-resolution points within a controllable area, with adjustable width limited to keep rendering within the performance threshold.Displayed trees are tinted in context to indicate assessment, while selected trees receive a colour multiplier for annotation.Source: §4.2.3 · p. 5

4.2.4 Techniques

  • Wedgelight uses a controller-aimed wedge; Remote Lantern moves a fixed cylindrical focus relative to the frame-of-reference; Ray Lantern aims the cylinder with a controller ray.The cylinder was selected for stem assessment because it shows full tree height and avoids obfuscating trees away from the user.Source: §4.2.4 · p. 5

4.3 Implementation

  • The Unity implementation filters mesh fragments using a focus transformation matrix, drawing full-resolution points inside the focus and synthetic context outside it.The implementation uses a high-resolution forest point cloud and FSCT-processed context; materials also render selection states in different colours.Source: §4.3 · pp. 5–6, §4.2.2 · p. 5

5 STUDY 2: VR F+C TECHNIQUES

The summative study evaluated three Focus+Context techniques against a control using a tree-marking task analogous to forestry operations. It varied technique and grid size and measured identification, movement, effort, usability, and preferences.

5.0.1 Task

  • Participants marked double-leader trees and distinguished individual trees from adjacent pairs during timed 60-second forest-grid tasks.The task was presented as a simplified analogue of tree selection in forestry.Source: §5.0.1 · p. 6

5.0.2 Study 2 Dataset Generation

  • A generator combined scans of 14 real trees with false-positive and false-negative variants, distributing double leaders across grids with randomized scale, rotation, and position.The source scans included 10 single leaders and 4 double leaders; the generator produced 1×10228 possible permutations for each task.Source: §5.0.2 · p. 6

5.0.3 Control

  • The control toggled between full context and full focus, but used subsampled point clouds because rendering the full-resolution grid exceeded performance limits.The setup was designed to remain at or above 70 fps, with large grids fixed at 15 × 15 trees.Source: §5.0.3 · p. 6

5.0.4 Independent Variables

  • The independent variables were interaction technique—Wedgelight, Remote Lantern, Ray Lantern, or Control—and forest-grid size.Conditions were counterbalanced to reduce fatigue or learning effects.Source: §5.0.4 · p. 6

5.0.5 Dependent Variables

  • Dependent measures covered tree identification and error, cluster identification, user and focus movement, frame-of-reference transformation, effort, usability, and preference.Movement measures were recorded at 0.1-second intervals; effort ratings followed each condition.Source: §5.0.5 · p. 6

5.0.6 Hypothesis

The study specified hypotheses about technique performance, grid size, and user preference, then used a standing VR protocol with repeated forest-grid tasks. Participants were university staff and students.

5.0.6 Hypothesis

  • The hypotheses predicted better performance with the three Focus+Context techniques than Control, an effect of grid size, preference for Focus+Context, and a technique effectiveness ranking.Source: §5.0.6 · p. 6

5.0.7 Apparatus

  • The study was conducted standing in an unobstructed 3 m2 × 3 m2 area using the VR headset and tethered laptop setup.Source: §5.0.7 · p. 6

5.0.8 Procedure

  • Participants trained on a 5×5 grid, then completed five 15×15 and five 10 × 10 grid tasks per technique in randomized order, with rest breaks available.There were four technique conditions.Source: §5.0.8 · p. 6

5.0.9 Participants

  • Twenty-four university staff and students aged 20–45 participated; 16 were male and 8 female.Participants had normal or corrected-to-normal vision.Source: §5.0.9 · pp. 6–7

5.1 Study 2 Results

The F+C techniques improved tree identification and reduced tagging errors relative to Control. Participant movement and focus interaction differed across techniques, while the study found no interaction between grid size and other measured factors.

5.1 Study 2 Results

  • Grid size had no interaction effect with other factors on any measure.The analysis treated identification percentages as fractions of the whole because the grid sizes had different totals.Source: §5.1 · p. 7

5.1.1 Tree Identification and Error

  • Control identified fewer double-leader trees (M = 2.70, SD = 1.97) and clusters (M = 1.55, SD = 1.45) than Wedgelight, Ray Lantern, or Remote Lantern.The respective tree-identification means were 6.24, 6.5, and 6.54; cluster-identification means were 4.12, 4.20, and 4.50.Source: §5.1.1 · p. 7
  • Control had more incorrect tree tagging (M = 3.35, SD = 4.79) than Wedgelight (M = 0.383, SD = 1.10), Ray Lantern, or Remote Lantern.Ray Lantern had M = 0.533, SD = 1.65; Remote Lantern had M = 0.371, SD = 1.29.Source: §5.1.1 · p. 7

5.1.2 Participant Movement

  • Control and Remote Lantern involved more headset movement than the egocentric techniques; Control also had the highest eye rotation, while Wedgelight had the lowest headset rotation.Headset movement means were 13.5m for Control and 13.3m for Remote Lantern versus 11.3m for Wedgelight and 11.9m for Ray Lantern; eye rotation ranged from 412.2rad for Control to 306.0rad for Wedgelight.Source: §5.1.2 · p. 7

5.1.3 Interaction Technique Usage

  • Wedgelight and Remote Lantern required less focus movement than Ray Lantern, while Wedgelight involved more focus scaling than Ray Lantern or Remote Lantern.Focus movement means were 11.9m for Wedgelight, 10.9m for Remote Lantern, and 20.0 for Ray Lantern; Wedgelight focus scale was M = 4.150 versus Ray Lantern M = 0.177.Source: §5.1.3 · p. 7
  • Control involved more frame-of-reference rotation and scaling than each F+C technique.Control rotation was M = 11.6rad and scale was M = 2.02; F+C rotation means ranged from 3.74rad to 5.75rad, and scale means were 1.19 to 1.21.Source: §5.1.3 · pp. 7–8

5.2 Participant Feedback

Participants reported lower mental effort and higher usability for the F+C techniques than for Control. Preferences also favored particular techniques for several measures, without establishing a single overall winner.

Participant Feedback

  • Control required higher mental effort (M = 5.46, SD = 1.32) than Wedgelight, Ray Lantern, or Remote Lantern, while physical effort did not differ significantly by technique.The mental-effort means were 3.79, 3.26, and 3.36, respectively; physical effort had χ2(3) = 5.887, p = 0.117.Source: §5.2 · p. 8
  • All three F+C techniques scored higher in usability than Control, which was rated ‘Poor’; Remote Lantern was ‘Excellent’ while the others were ‘Good’.System Usability means were 76.9 for Wedgelight, 78.8 for Ray Lantern, 81.0 for Remote Lantern, and 56.4 for Control.Source: §5.2 · p. 8
  • Remote Lantern and Ray Lantern were preferred over Control for accuracy, while overall preference favored both lantern techniques and enjoyment favored Remote Lantern and Wedgelight.Ray Lantern was also preferred over Control for assessment; the excerpt reports no overall ranking among the F+C techniques.Source: §5.2 · p. 8

5.3 Discussion of Study 2 Results

The F+C techniques outperformed Control on every performance metric, with fewer errors contributing to higher overall accuracy. The discussion also identifies task-specific preferences, possible extensions, and limits on the study’s comparison and scope.

5.3 Discussion of Study 2 Results

  • F+C techniques outperformed Control on every performance metric, with higher successful tree identification and marking and fewer errors contributing to higher overall accuracy.The authors suggest the techniques are advantageous for preventing false positives in tree identification.Source: §5.3 · p. 8
  • Grid size showed no evidence of affecting F+C performance; the authors frame assessment by the proportion of trees to be thinned rather than total forest area.This result contradicted hypothesis H2.Source: §5.3 · p. 8
  • Users preferred F+C for tree marking, but the study found no clear ranking among its techniques; lantern focuses suited the task, while Wedgelight was rated highest for enjoyment.The authors suggest Wedgelight may excel at finer tree-structure analysis.Source: §5.3 · p. 8
  • Participants suggested adjustable focus shapes, illumination highlighting, and additional controls for context height, individual tree rotation, and focus-tree scaling.Source: §5.3 · p. 8
  • The techniques may transfer to other fields, but those domains require alternate methods of creating context.Suggested applications include construction geometry inspection and validation of abstractions in archaeological and agricultural analysis.Source: §5.3 · p. 8
  • The Control condition used lower-resolution data; the authors call for direct comparison when VR performance is on par with remote-sensing fidelity.They also identify unbalanced participant gender and the need to expand evaluation to other forestry task analogues, such as species and age classes.Source: §5.3 · p. 8

6 CONCLUSIONS

The formative study found faster and more accurate forestry analytics with VR than desktop. The subsequent F+C study found the techniques performed better than Control, with focus form suited to different analysis needs.

6 CONCLUSIONS

  • Users were faster and more accurate with VR than desktop in the formative forestry analytics study.The authors associate VR’s depth perception and freedom of movement with structural analysis of dense point clouds, and report less controller movement than on desktop.Source: §6 · p. 8
  • All F+C techniques performed better than Control in the follow-up study, and the authors consider them viable for other fields with similar spatial data.Study data, point clouds, and scripts were shared for future research.Source: §6 · p. 8
  • Lanterns’ cylindrical focus suits comparative analysis of multiple trees, whereas Wedgelight’s directed focus may better support detailed structural analytics.The authors recommend exploring other context-generation methods and diverse synthetic datasets.Source: §6 · p. 8
Loading 2610.22878v1…