WristSketcher: AR Wristband Sketching

• WristSketcher is an augmented reality sketching system that uses a multilayer sensing wristband for accurate, low-fatigue 2D drawing.
• It employs a robust gesture recognition pipeline and dynamic animation framework to enable real-time AR content creation.
• User studies demonstrate high precision and comfort compared to mid-air gesture systems, highlighting its ergonomic advantages.

WristSketcher is an augmented reality (AR) content creation system designed around a flexible, multilayer sensing wristband. Its principal aim is to facilitate accurate, low-fatigue 2D sketching and dynamic asset creation for AR glasses in real-world scenarios where traditional sketching modalities—tablets and mid-air gestures—are either cumbersome or imprecise. WristSketcher systematically redefines the sketching interaction paradigm by shifting input from physically and socially problematic mid-air gestures to the compact and stable surface of the user’s wrist. The result is a zero-burden authoring workflow with high user satisfaction, precision, and comfort, as demonstrated through empirical user studies (Ying et al., 2022).

1. System Architecture and Sensing Principles

WristSketcher hardware comprises an 8.36 cm × 8.36 cm, 0.2 mm-thick multilayer pressure-sensitive wristband arranged as a 40 × 40 active point matrix (1.6 mm pitch), offering 1,936 independently addressable sensing locations. Pressure is detected via two polyester film layers separated by dielectric tape, which close circuit at contact. Signal acquisition is handled by a USB-connected module sampling the matrix at 60 Hz; the minimum detectable force is 20 g, with a response time of 10 µs.

Each frame yields a pressure matrix $P_k(i,j)$, which undergoes thresholding ($P_k(i,j) > T_{thr}$), 2D median filtering, and single-pass connected-component labeling. Clusters of at least five pixels are treated as potential finger contacts, with the touch-point per group $G$ taken at the location of maximal pressure $(i^*, j^*) = \arg\max_{(i, j) \in G} P_k(i, j)$. Sub-threshold or isolated clusters are filtered as noise. The processed data feeds subsequent gesture recognition and rendering modules (Ying et al., 2022).

2. Gesture Recognition and Interaction Design

The gesture recognition pipeline operates at approximately 31 FPS on a consumer laptop. For each frame, raw pressure matrices undergo preprocessing as above. The resulting touch points (typically one or two per frame) are aggregated over a 3-frame temporal window to stabilize detection. The system's time-based classifier then labels gestures as follows:

• Tap: Duration $\Delta t < 0.15$ s, and no second touch within 0.5 s.
• Double-Tap: Two taps within 0.5 s.
• Long-Press: $\Delta t > 1$ s.

User-defined heuristics, obtained from a 26-participant study, determined preferred mappings:

Command	Gesture
Main/Sub-menu	1-Finger Long-Press
Select Menu Item	Slide after Long-Press
Confirm	1-Finger Double-Tap
Undo	2-Finger Tap

This gesture vocabulary capitalizes on stability and ergonomic factors, favoring single or two-finger interactions over complex multi-finger or mid-air poses. This approach demonstrably reduces arm fatigue and social awkwardness compared to prior AR sketching paradigms (Ying et al., 2022).

3. Dynamic Sketch and Animation Framework

Assets within WristSketcher are represented as structs containing a polyline (geometry), visual parameters (color, thickness), and an animation list. Animation effects are authored via menu and evaluated each render frame. The primary animation types include:

1. Doodle: Each vertex is jittered per-frame with $\delta x, \delta y \sim \mathrm{Uniform}(-A, A)$.
2. Frame: User-sketched frames are cycled temporally by $f(t) = \lfloor (t - t_0) \cdot r \rfloor \bmod M$.
3. Emit: Assets are treated as particles; user-sketched lines define the emitter and initial velocity $\vec v$, subject to gravity $\vec g$:

$P_k(i,j) > T_{thr}$0

1. Rotate: Asset rotates about centroid $P_k(i,j) > T_{thr}$1 with angular rate $P_k(i,j) > T_{thr}$2, so angle $P_k(i,j) > T_{thr}$3.
2. Move: Linear interpolation along a user-defined polyline path.

Multiple effects may be bound simultaneously to a single asset, supporting complex animations with minimal interaction overhead.

4. Quantitative Evaluation and User Study

Performance evaluation involved both objective metrics and subjective user studies:

• Gesture Recognition: Overall accuracy reached 96.0% (SD=2%), with the highest accuracy for one-finger long-press (97.3%) and lowest for one-finger double-tap (92.0%); confusions typically arose at low pressures or ambiguous timings.
• Sketch Precision: Compared to freehand mid-air sketching using AR glasses, WristSketcher yielded significantly lower drawing error (measured as mean pointwise deviation from template shapes: rectangle, triangle, circle), with $P_k(i,j) > T_{thr}$4. Completion was slower ($P_k(i,j) > T_{thr}$5), but users reliably traded off speed for superior accuracy.
• User Satisfaction: In a study with $P_k(i,j) > T_{thr}$6, mean Likert scores were: overall 4.01, ease of use 4.10, usability 4.03, functionality 3.90. Qualitative responses highlighted comfort, learnability, and precision; feedback suggested enhancements in haptic/auditory feedback and menu navigation (Ying et al., 2022).

Pre-defined animation tasks (e.g., twinkling stars, rain) averaged ~3.8 min creation time; free-form sketches included clothing art, memory aids, and playful AR scenes.

5. Context: Comparative Modalities and Related Work

WristSketcher addresses ergonomic and usability challenges observed in prior systems. Mobile devices offer high accuracy but are encumbering, while mid-air gestures suffer from arm instability, fatigue, and social acceptability issues (Ying et al., 2022). StripBrush, a related VR sketching interface, demonstrated that relaxing brush orientation constraints minimizes wrist motion and perceived physical demand ($P_k(i,j) > T_{thr}$7 NASA-TLX; $P_k(i,j) > T_{thr}$8 SUS usability), while increasing drawing accuracy ($P_k(i,j) > T_{thr}$9) (Rosales et al., 2021). Haptic forearm feedback, as explored in Sarac et al., further evidences the potential of wrist- and arm-based modalities for expressive and believable user interaction (Sarac et al., 2019).

WristSketcher's surface-based sketching represents a distinct design choice, providing a locally stable, low-effort interaction zone with theoretically lower biomechanical strain. A plausible implication is that further integration of haptic feedback (normal force stimulation) could reinforce the believability of AR content manipulation, following recommendations from haptic sketch studies (Sarac et al., 2019).

6. Applications, Limitations, and Future Directions

WristSketcher has demonstrated utility for art creation (animated T-shirt motifs, decorative AR "lanterns"), information aids (garbage sorting instructions), and entertainment (virtual weather, cartoon sequences). However, it is currently limited to 2D content due to the absence of simultaneous localization and mapping (SLAM) capability on the host AR glasses, precluding robust 3D anchoring and persistence. Animated effects in the present implementation are pre-scripted and do not respond to real-world events or user motion (Ying et al., 2022).

Proposed future directions include:

• Integrating SLAM for live 3D anchoring and multi-view consistency.
• Augmenting feedback channels (haptic, audio) for richer interaction.
• Advancing gesture recognition via on-device machine learning to expand the interactive vocabulary.
• Enabling responsive, physics-driven dynamic effects governed by scene and user context.

7. Synthesis and Design Insights

Surface-based sketching via wristband offers a stable, precise, and low-fatigue alternative to mid-air gesture input, with high recognition accuracy and user acceptance. Constraints on interaction area (wrist surface) naturally reduce operational speed compared to freehand sketching, but this is offset by increased precision and user comfort. The system leverages a reduced, pragmatically chosen gesture vocabulary—tap, double-tap, long-press—resulting in efficient menu and canvas manipulation. Embedding dynamic animation effects directly within AR sketching workflows broadens the scope of creative and communicative possibilities, bridging static illustration and time-varying storytelling (Ying et al., 2022).

WristSketcher's architecture and methodologies reflect a convergence of findings from ergonomics, pressure-based sensing, lightweight AR, and user-centered interaction design. The approach stands in methodological and empirical contrast to VR brush-based techniques and forearm haptic feedback explorations, but shares with them a commitment to reducing effort while maximizing expressivity and user immersion (Rosales et al., 2021, Sarac et al., 2019).