Human Activity Recognition of Figure Skating Jumps Using Abdominal Electromyography and Accelerometry

Figure skating is a sport where split-second decisions and precise body control determine success. A skater must launch into the air, spin multiple times, and land on a thin blade of ice without faltering. For decades, coaches have relied on their own eyes and slow-motion video to judge these jumps, looking for subtle errors in takeoff or landing. However, cameras have a blind spot: they can only see what is on the outside. They cannot see the muscles firing inside the body to prepare for the jump, nor can they always see clearly when a skater's body twists so fast that limbs block the view. To solve this, scientists have turned to wearable sensors, small devices that stick to the skin to record movement and muscle activity, offering a new way to understand the hidden mechanics of athletic performance.

A team of researchers at the Technical University of Ostrava and the University of Ostrava in the Czech Republic has developed a new method to automatically identify specific figure skating jumps using these wearable sensors. Their goal was to create a system that could tell the difference between six different types of jumps, ranging from single rotations to complex triple rotations, without needing a camera or a team of judges. Instead of watching the skater, their system listens to the body's internal signals. They attached small sensors to the backs of twelve elite female adolescent skaters. One sensor measured the acceleration of the body in three directions, while two other sensors placed on the abdomen measured the electrical activity of the core muscles. This combination allowed the researchers to capture both the physical movement of the jump and the muscle preparation that happens just before the skater leaves the ice.

The researchers collected data from hundreds of jump attempts, including single, double, and triple versions of the Axel, Salchow, and Loop jumps. They then fed this information into a computer system designed to learn patterns. The team tested several different types of computer learning methods to see which one could best sort the jumps into the correct categories. They found that the most successful approach was not a complex, deep-learning network often used for image recognition, but rather a simpler, tree-based method that made decisions by following a series of logical rules. This system achieved a remarkable level of accuracy, correctly identifying the type of jump nearly 99 percent of the time. The system worked by looking at short, specific slices of time from the sensor data, finding that a window of data covering 0.6 seconds with a small overlap worked best to capture the unique signature of each jump.

A key discovery in this study was understanding which parts of the sensor data were actually necessary for the computer to make the right choice. The researchers ran tests where they removed certain sensors to see if the system would still work. They found that while the movement sensor alone could do a decent job, the system performed significantly worse if they removed the sensors measuring the abdominal muscles. This confirmed that the way a skater tightens their core muscles before jumping is a critical clue that distinguishes one jump from another. Furthermore, they discovered that the side-to-side and up-and-down movements of the body were the most important physical signals, while the forward-and-backward movement mattered much less. This suggests that the specific lean a skater uses to start a turn and the force with which they hit the ice upon landing are the primary physical markers that define a jump.

The study also addressed a common problem in sports technology: the need for bulky equipment. Many previous attempts to track skating jumps required multiple sensors placed on the wrists, ankles, and waist, which can be heavy and distracting for an athlete. This new method proved that a single sensor on the back, combined with two small sensors on the abdomen, was enough to achieve top-tier results. This lighter setup means the system is less likely to interfere with the skater's natural movement, making it more practical for real-world training. While the current study focused on successful jumps performed by young female skaters in a controlled setting, the researchers believe this approach lays the groundwork for future tools that could help coaches analyze jump quality in real time. By identifying the exact moment a jump is executed and classifying it instantly, such technology could eventually help athletes refine their technique and prevent injuries, all while remaining invisible to the naked eye.