Endpoint Distribution Modeling-Based Capture Algorithm for Interfering Multi-Target

Xiangliang Zhang, Junlin Li, Pengjie Li, Fang Si, Xiangzhi Liu, Yu Gu, Shuguang Meng, Jibin Yin, Tao LiuSensors 2024 · DOI

TL;DR

This paper introduces novel models (ST-PEDM and MT-PEDM) and algorithms (BRPA and ERPA) to improve the accuracy and efficiency of pointing interactions in physical spaces, especially when dealing with multiple interfering targets. These advancements significantly boost target selection accuracy, paving the way for more intuitive control of devices like remote controls and drone swarms.

What Problem Does This Paper Address?

In virtual environments, pointing interactions are often aided by visual cues like cursors or rays. However, in physical spaces, users must rely solely on their perception and experience to accurately select targets. This is challenging due to factors like hand tremors, physiological limitations, and the "Heisenberg effect" (where the act of confirming a selection can alter the pointing direction). Existing methods for modeling target distribution and partitioning interaction areas in physical space are often simplistic and lack theoretical rigor, especially when multiple targets are present and can interfere with each other. This paper aims to bridge this gap by developing more robust models and algorithms for accurate and efficient target capture in complex physical interaction scenarios.

Methods

The core of this research lies in developing two endpoint distribution models:

  • Single-Target Pointing Endpoint Distribution Model (ST-PEDM): This model accounts for factors like target distance, visual error, and the Heisenberg effect to predict the distribution of pointing endpoints for a single target. It builds upon the concept of a two-dimensional normal distribution, incorporating parameters influenced by these physical interaction challenges.
  • Multi-Target Pointing Endpoint Distribution Model (MT-PEDM): This model extends the ST-PEDM to handle scenarios with multiple targets, specifically considering the interference between targets. It models how the presence and relative positions of interfering targets affect the endpoint distribution of the intended target.

Based on these models, two region partitioning algorithms are proposed:

  • Basic Region Partitioning Algorithm (BRPA): This algorithm is designed for simpler scenarios with a single target and no interference. It partitions the space based on the ST-PEDM.
  • Enhanced Region Partitioning Algorithm (ERPA): This algorithm is more sophisticated and handles complex multi-target scenarios with interference. It uses Voronoi diagrams to divide overlapping regions when interfering targets are close, ensuring each target has a distinct interaction zone.

Datasets

The study utilized experimental data collected from 15 participants performing pointing tasks. No pre-existing datasets were used for training or evaluation. Instead, a custom experimental setup was designed and implemented to gather data on pointing accuracy and task completion times under various conditions (different target distances, presence of interfering targets, and Heisenberg effect manipulation).

Key Insights and Ideas

  • The Heisenberg Effect is Quantifiable: The study demonstrates that the Heisenberg effect, a known phenomenon in human-computer interaction, significantly impacts pointing accuracy and can be modeled.
  • Interfering Targets are a Major Challenge: The presence of nearby targets substantially degrades pointing accuracy, and their orientation and distance play crucial roles in this degradation.
  • Voronoi Diagrams for Precise Partitioning: The ERPA's use of Voronoi diagrams proves effective in mathematically dividing complex, overlapping interaction regions, leading to more accurate user intention detection.
  • Physical Space Demands Different Models: The proposed models (ST-PEDM and MT-PEDM) are tailored for physical pointing interactions, acknowledging limitations not present in virtual environments.

Key Results

The proposed algorithms significantly improved pointing accuracy and efficiency:

  • Accuracy Improvement:
    • At 200 cm target distance, accuracy increased from 60.54% (no algorithm) to 72.94% (BRPA) and 84.11% (ERPA).
  • At 300 cm target distance, accuracy increased from 42.39% (no algorithm) to 68.57% (BRPA) and 82.74% (ERPA).
  • Efficiency Improvement: For target distances of 200 cm and 300 cm, task completion times were reduced by half or more with BRPA and ERPA compared to no algorithm. ERPA was also faster than BRPA in directional selection tasks.
  • Model Fit: The proposed ST-PEDM showed a better fit to the experimental data compared to the existing EDmodel for cursor-free pointing tasks.

Limitations and Future Work

  • Laboratory Setting: The experimental validation was conducted in a controlled laboratory environment using high-precision motion capture. Real-world scenarios may involve less accurate sensors (e.g., ultra-wideband technology with sub-meter distance and 15-degree angular errors), which could affect the model's performance.
  • Future Work: The authors plan to:
  • Validate the proposed models and algorithms in real-world scenarios using less precise sensor technology.
  • Apply the approach to rapid capture control and trajectory planning for drone swarms, integrating it with smart glasses for command and data transmission.

Summary updated

Abstract

In physical spaces, pointing interactions cannot rely on cursors, rays, or virtual hands for feedback as in virtual environments; users must rely solely on their perception and experience to capture targets. Currently, research on modeling target distribution for pointing interactions in physical space is relatively sparse. Area division is typically simplistic, and theoretical models are lacking. To address this issue, we propose two models for target distribution in physical space-pointing interactions: the single-target pointing endpoint distribution model (ST-PEDM) and the multi-target pointing endpoint distribution model (MT-PEDM). Based on these models, we have developed a basic region partitioning algorithm (BRPA) and an enhanced region partitioning algorithm (ERPA). We conducted experiments with 15 participants (11 males, and four females) to validate the proposed distribution models and region partitioning algorithm. The results indicate that these target distribution models accurately describe the distribution areas of targets, and the region partitioning algorithm demonstrates high precision and efficiency in determining user intentions during pointing interactions. At target distances of 200 cm and 300 cm, the accuracy without any algorithm is 60.54% and 42.39%, respectively. Using the BRPA algorithm, the accuracy is 72.94% and 68.57%, while, with the ERPA algorithm, the accuracy reaches 84.11% and 82.74%, respectively. This technology can be utilized in interaction scenarios involving handheld pointing devices, such as handheld remote controls. Additionally, it can be applied to the rapid capture control and trajectory planning of drone swarms. Users can quickly and accurately capture and control target drones using pointing interaction technology, issue commands, and transmit data through smart glasses, thereby achieving effective drone control and trajectory planning.
BibTeX
@inproceedings{zhang2024endpoint,
  author = {Xiangliang Zhang and Junlin Li and Pengjie Li and Fang Si and Xiangzhi Liu and Yu Gu and Shuguang Meng and Jibin Yin and Tao Liu},
  title = {{Endpoint Distribution Modeling-Based Capture Algorithm for Interfering Multi-Target}},
  booktitle = {Sensors},
  year = {2024},
  doi = {10.3390/s24248191},
}
Figure