Quality of tracking methods with object reidentification

Objective. This paper presents a study of object tracking algorithms for analyzing sports competitions with a minimum permissible number of frames per second without critical loss of quality of operational systems. The purpose of the study is to determine the optimal tracking method and reidentification model with an increase in the time interval between frames. Method. The study used the Re-Identification MSMT17, Market1501, DukeMTMCreID datasets; specialized neural networks OsNet, ResNet and MobileNet. Tracking algorithms BotSort, Bytetrack, OcSort, DeepOcSort, StrongSort. The NOTA and IDF1 quality metrics were used. Result. The metrics of the experiments are generally quite low, which is due to the nonlinearity of athletes' movement, unlike pedestrians' movement, and a large number of changes in their posture. Conclusion. Depending on the degree of information compression, the optimal tracking methods when using pre-trained reidentification models are ByteTrack and OcSort.