Evaluating OCR Performance for Assistive Technology: Effects of Walking Speed, Camera Placement, and Camera Type

TL;DR

The paper evaluates OCR performance for assistive technology under static and dynamic conditions, focusing on walking speed, camera placement, and camera type. It employs a controlled experimental framework with signage of varied fonts and backgrounds, three mounting positions (shoulder, head, hand), two phone cameras (main and ultra-wide), and Meta glasses, across four OCR engines (Google Vision, PaddleOCR, EasyOCR, Tesseract). Key findings show Google Vision achieving the highest accuracy overall, PaddleOCR as the strongest open-source alternative, and the main camera outperforming the ultra-wide while shoulder mounting provides modest gains; accuracy declines with speed and viewing angle, with Meta glasses yielding substantially lower performance. These results inform practical design trade-offs and suggest adaptive sensing strategies and on-device OCR to support real-world navigation and reading for people with visual impairment.

Abstract

Optical character recognition (OCR), which converts printed or handwritten text into machine-readable form, is widely used in assistive technology for people with blindness and low vision. Yet, most evaluations rely on static datasets that do not reflect the challenges of mobile use. In this study, we systematically evaluated OCR performance under both static and dynamic conditions. Static tests measured detection range across distances of 1-7 meters and viewing angles of 0-75 degrees horizontally. Dynamic tests examined the impact of motion by varying walking speed from slow (0.8 m/s) to very fast (1.8 m/s) and comparing three camera mounting positions: head-mounted, shoulder-mounted, and hand-held. We evaluated both a smartphone and smart glasses, using the phone's main and ultra-wide cameras. Four OCR engines were benchmarked to assess accuracy at different distances and viewing angles: Google Vision, PaddleOCR 3.0, EasyOCR, and Tesseract. PaddleOCR 3.0 was then used to evaluate accuracy at different walking speeds. Accuracy was computed at the character level using the Levenshtein ratio against manually defined ground truth. Results showed that recognition accuracy declined with increased walking speed and wider viewing angles. Google Vision achieved the highest overall accuracy, with PaddleOCR close behind as the strongest open-source alternative. Across devices, the phone's main camera achieved the highest accuracy, and a shoulder-mounted placement yielded the highest average among body positions; however, differences among shoulder, head, and hand were not statistically significant.

Figures (14)

• Figure 1: A photo of signage used in the study, illustrating variation in font size, color, and background contrast.
• Figure 2: Camera mounting positions evaluated in the study: (a) shoulder-mounted, with the smartphone attached to a backpack strap; (b) head-mounted, with the smartphone secured to a headband; and (c) hand-held, with the smartphone held naturally in front of the body.
• Figure 3: Static test setup for evaluating OCR detection range. (a) Diagram of the test field showing camera placement and sign-board positions at varying distances and angles (red dots); (b) sign board at low height (33 cm); (c) sign board at medium height (100 cm); and (d) sign board at high height (218 cm).
• Figure 4: Dynamic test setup. Dynamic test setup. A 7 m straight walkway was marked between a start line and a finish line. The participant walked along this line at four controlled speeds. The sign board was placed so that the line from the start point to the sign subtended an azimuth angle of 0°, 15°, or 30° relative to the walking direction, while the participant’s path remained straight. Thus, at 0° the participant walked directly toward the sign, and at 15° and 30° the sign was laterally offset from the path.
• Figure 5: Heatmaps of average OCR accuracy for the main camera at the high, medium, low heights from top to bottom. Each panel shows character-level accuracy for one engine (PaddleOCR, Tesseract, EasyOCR, Google Vision) across distances (1–7 m, y-axis) and viewing angles (0°, 15°, 30°, x-axis). Brighter colors indicate higher accuracy.