Computer vision, even 3-D vision, is not completely accurate.

Judge as of July 2018. True: the evidence shows that computer vision systems, including 3-D vision such as stereo, depth sensing or lidar-based perception, had non-zero error rates in practice and on benchmarks, i.e. no system was perfectly accurate. Mostly true: this holds in general, but the evidence shows some narrow, controlled 3-D vision tasks at effectively perfect accuracy. Mixed: the evidence is split on whether 3-D vision in particular was accurate enough to count as complete in common uses. Mostly false: credible sources show that 2-D or 3-D vision had in fact been made completely accurate in broad classes of tasks. False: computer vision, including 3-D vision, was generally shown to be completely accurate.

Set before the research began.

1. E1 Error Analysis in a Stereo Vision-Based Pedestrian Detection Sensor for Collision Avoidance Applications PMC (National Library of Medicine), peer-reviewed journal article · 13 April 2010

   “For example, at 15 m the depth error is about ±1.5 m, at 10 m is about ±0.7 m and at 5 m is lower than ±0.2 m.”

2. E2 Obstacle Classification and 3D Measurement in Unstructured Environments Based on ToF Cameras PMC (National Library of Medicine), peer-reviewed journal article · 14 February 2014

   “As shown in Table 10, the recognition accuracy of obstacles reaches 93%.”

3. E3 Obstacle Classification and 3D Measurement in Unstructured Environments Based on ToF Cameras PMC (National Library of Medicine), peer-reviewed journal article · 14 February 2014

   “Secondly, the relative error illustrated in Tables 2, 3, 4, 5 and 6 is usually lower than 5% except for some special overlapped or very irregular cases in Tables 2 and 3.”

4. E4 Google develops computer vision accurate enough to solve its own CAPTCHAs The Verge · 16 April 2014

   “Despite being near perfect when it comes to CAPTCHAs — a feat that plenty of humans can’t even manage — the new system’s analysis of Street View imagery isn’t quite as accurate, correctly identifying the text just over 90 percent of the time.”

Peer-reviewed studies from before July 2018 report measurable error in 3-D vision. A stereo-vision pedestrian sensor had depth errors of about ±1.5 m at 15 m [E1]. A time-of-flight camera system reached 93% obstacle recognition accuracy, with relative errors usually below 5% and larger errors in some overlapped or irregular cases [E2, E3]. For 2-D vision, Google's text recognition on Street View imagery was correct just over 90% of the time [E4]. None of the evidence shows a computer vision system that was completely accurate, so the claim is resolved true.