Nowadays, computer vision is a common tool for video analysis. Many research groups have put significant amount of effort to improve robustness in these applications for years. We hypothesized that the video input itself played an important part in the robustness issue. More specifically, inconsistency of frame rate negatively affects the accuracy in time-related applications. To validate the hypothesis, we explored how common video-capturing methods performed regarding frame-rate consistency and analyzed how it affected performance in image-based measurements of the heart rate and object velocity. We found that common video capturing methods failed to create videos with a regular frame rate. Yet, it could be hard to perceive by human observers and we may process a video without awareness of the issue. This inherently lowers the accuracy of certain applications. To ameliorate the problem, this work proposes a video-capturing method that carefully controls the pace of video frames; thereby, making the intervals of video frames virtually equal as intended. The main idea of this method is to postpone I/O and encoding tasks until recording is finished through initially storing raw video frames in the memory and pacing frame capturing according to their ideal timing. Experiments demonstrated that the proposed video frame-pacing method had significant impacts on accuracy of time-related applications. In object-velocity measurement, typical video-capturing methods had error approximately 3 to 5%, while our method had error less than 2%. In non-contact heart-rate measurement, typical methods had error about 6.8616 to 19.0021 bpm, while ours had error merely 0.7548 bpm.