I have been working with ZED camera recorded videos recently. Every time I load an SVO2 file into the Explorer program, it takes 1–2 minutes to process/play a 50-second video. The processing time increases with the size of the video. Is this expected behavior, or am I missing something?
From the documentation I found, the examples focus on real-time recording, but this approach doesn’t seem to work as effectively for recorded videos. Specifically, when running the custom detector (custom detector), I am unsure whether working with recorded videos is better or if real-time processing is preferable.
If my requirement is to work primarily with recorded videos, what would you recommend as the best approach?
Additionally, what is the recommended speed at which the camera should be mounted on a moving vehicle to achieve accurate localization of detected objects, both in real-time and during post-recording analysis back in the office?
That is correct, there is a small overhead time to load a video depending on its size (i.e. duration, compression mode used, bitrate, …). For example, videos compressed with H265 compression will have a smaller size than lossless compression, and therefore will be faster to load.
Regarding real-time vs offline processing, this really depends on your use case. If your application makes it possible to process SVOs offline, this would be recommended as you can use the more compute-demanding algorithms, with the best detection results. If your application needs to run real-time on the vehicle, you may have to use more compute-efficient algorithms that may deliver reduced performance in comparison.
I’m not sure to have understood the question. If you are asking about the frame rate of the application, I would say the highest would be the best for a moving vehicle, to reduce motion blur to a maximum.