The athletes made brilliant achievements in the skiing field, which also attracted the attention of a virtual coach named “Guanjun”.
In fact, “Guanjun” is not human, but the artificial intelligence referee and coach system developed by Xiaobing company. Before the Winter Olympics, “xiaoice International Atomic scoring system” (x-iass) has silently served the freestyle skiing aerial skills team for more than three years.
In 2018, Guan Jun began to learn and master the relevant knowledge of air skills projects, and joined the national air skills team in 2019, hoping to improve the training efficiency and effect of Freestyle Skiing air skills projects with the help of artificial intelligence.
To achieve the above objectives, Guanjun Xiaobing system has to achieve several key points: first, rapid scoring, second, quantitative action, and third, the traceability and predictability of training data. Due to the danger of air skills, athletes have limited training every day. In order to improve training efficiency, it is necessary to strengthen athletes’ memory of high score movements. Guanjun needs to make professional judgment for each jump of athletes, keep a high degree of consistency with the scoring criteria of international judges, strictly judge the score deduction actions, and quantify the whole process actions in the three stages of take-off, air and landing, which can support the multi-dimensional index analysis of motion trajectory, body posture, exit angle and height and distance, so as to provide guidance for coaches. At the same time, we need to be able to trace every jump of athletes, accumulate data for a long time, and build sports files for scientific observation and prediction of athletes’ performance.
Taking action recognition as an example, the speed of male athletes out of the landslide exceeds 70 kilometers per hour, the height difference is up to 15 meters, the stagnation time of the whole set of actions is only 2-3 seconds, and the most difficult air turning and turning action can reach 8 laps. At night, the picture of athletes is often integrated with the background, and the observation distance is as far as 25 meters, In addition, the severe cold of minus 20 or 30 degrees makes it difficult for high frame rate video capture equipment to work for a long time. It is a great challenge to build an artificial intelligence referee and coach system of “real-time scoring, quantitative action, traceability and prediction”.
To break through these challenges, Xiaobing team is also a “science and technology Olympiad”. The first is to solve the problem of posture recognition: the first step is to overcome the interference of complex background in large scenes and realize the accurate recognition of athletes on the premise of long-distance shooting; Secondly, the self-developed target tracking algorithm realizes the accurate positioning of athletes in high-speed sports; More importantly, through continuous iterative self-developed posture recognition algorithm, the accurate recognition of athletes’ posture is finally realized, which ensures the accuracy of athletes’ action and posture discrimination. The second is how to build the analysis model when the training data is very scarce: Based on the Xiaobing framework, under the condition of small samples, Xiaobing’s unique data amplification technology is used to carry out the model training and data analysis practice of ice and snow sports scenes, and the xiaoice cvanalysis model for winter sports is self-developed and used for daily training.
Guan Jun, who has broken through many challenges, has lived up to expectations. As a virtual coach for three years, he has completed nearly 10000 scores and more than 50000 movement analysis to assist athletes in daily training efficiency and improve the effect. He has become an important assistant for coaches in daily teaching.
In the Beijing Winter Olympics test competition held in February last year, Guan Jun also served as the only competition referee for air technology projects, and successfully completed 44 person times of individual pre-finals, super finals and group pre-finals, which is also the first time that artificial intelligence has independently completed the task of competition execution in the world.
There are two preconditions for Xiaobing team to successfully build Guanjun. One is to continuously invest in the research and development of underlying technologies, including natural language processing, computer vision, computer voice, artificial intelligence creation and other technology stacks, so as to maintain the foresight and integrity of the technology; The second is to work with leading industry customers and partners to jointly create mature, complete and valuable virtual human landing applications.
Li Di, CEO of Xiaobing company, said: “Xiaobing will continue to work with customers and partners in various industries to create AI beings that are tireless, safe, reliable and stable output for hundreds of millions of users. In the field of sports, we plan to cover a complete scene from professional AI coaches in the arena to sports information anchors, virtual employees in event venues, campus sports assistants, mass fitness coaches and even AI designers of sports brands 。”
