DETERMINATION AND COMPARISON METHODS OF BODY POSITIONS ON STREAM VIDEO

Context. One of the tasks computer vision is task determining human body in image. There are many methods to solve this problem, some based on specific equipment (motion capture, kinect) and provide highest accuracy, give less accuracy but do not require additional use computing power. But usually, such has a high cost, so ensure low cost developments designed determine image, you should develop algorithms technology. These can then be applied various fields analyze compare positions for variety purposes.
 Objective. The aim work study effectiveness existing libraries position as well comparing obtained poses terms speed determination.
 Methods. A set pose comparison were analyzed purpose developing system correctness exercise by user real time. OpenPose, PoseNet BlazePose their suitability recognizing tracking parts movements real-time video streams. advantages disadvantages each library evaluated performance, computational efficiency. Additionally, different analyzed. algorithm was ability accurately positions.
 As result, combination weighted distance method achieve best performance recognition, with robustness across range challenging scenarios. further enhanced techniques L2 normalization alignment improve its generalization. Overall, offers powerful effective solution F1 index.
 Results. Existing models have shown similar results quality determination run-up about 2%. When cross-platform software product, library, which an API working directly browser mobile platforms, significant advantage accuracy. Also, uses extended 33 keypoint topology it becomes applicable wider list tasks. In methods, greatest influence exerted Conclusions. Among comparison, distances showed results. inversely proportional significantly exceeds recommended value – 40ms.