5.3.1. Heart Rate

We note that the heart-rate measurements were deemed to be the least accurate in our system. According to the responses of the follow-up questions in the post-run survey, the subjects found the pre- and post-run single heart-rate measurements using the computer vision algorithms to be inaccurate. This discontent can be explained by the fact that these algorithms were mainly tested in a controlled lab environment, and failed to perform at the same level when taken in uncontrolled settings. For example, on different occasions, subject p542ok responded to the follow-up questions with "I am not at 70 bpm immediately after a run [(referring to the vision-based estimates)]", "face and finger[-based estimates] are way off as usual", (they were lower than 55 bpm) "finger[-based estimates] is still way too low after the run" and "finger heart rate after run was 45". In the first instance reported

above, the heart-rate chest strap did indeed record 70 bpm at the end of the run. In all other instances when the subject complained about the accuracy of the vision-based heart-rate estimates, the heart-rate chest strap recorded values that were significantly higher than the vision-based ones. No feedback was given regarding the accuracy of the chest heart-rate strap.

On one occasion, the same subject took video-based heart rate estimates without actually running. The subject did not report any perceived inaccuracy in the estimates, which were 61 bpm and 62 bpm for the finger-video-based and face-video-based estimates, respectively. Other subjects did not provide textual feedback regarding the accuracy of the heart-rate measurements.
