In video analysis in swimming, athletes are usually recorded by multiple cameras and thus from multiple perspectives. For example, the multicamera system can be set up in such a way that there are several cameras installed on the side - at intervals of 0m (starting block), 2.5m, 5m, 10m and 15m both underwater and above water. This way the coach and athlete have 5 individual shots available after training.
The problem with this is that these 5 separate recordings have to be analyzed for the evaluation, while the athlete quickly "swims" through the individual videos, so he is only in the picture for a short time at a time.
An automatic camera merge is intended to remedy this situation. The various camera views are combined in a single shot so that the athlete can be seen swimming through the different images at a glance. The result is a fluid video that shows the complete immersion phase of the athlete.
For this intersection, the corresponding configuration must be made for a specific camera setting. In the process, the interfaces of the image sections are put together "by hand", similar to a jigsaw puzzle. Then the pixel coordinates of the upper left edges are calculated.
After that, a specially developed algorithm and the libraries of the FFmpeg Foundation are used. The coordinates are used to merge the videos appropriately. In this way, a whole is created from several videos.
However, cropping also has a disadvantage: there is a bend in the shot at the intersection of the images, caused by the perspective distortion. The further away the athlete is from the camera, the greater the offset.
To keep the offset as small as possible, the pixel coordinates for each path should be specified. This way, the merge can be optimized accordingly. Before recording, the mode must then be selected so that the videos can be laid end to end accordingly.