Introducing SKDMap-Net: A New Gait Analysis System
On July 8, 2026, at 16:28, a study unveiled SKDMap-Net, a novel system capable of analyzing human gait for identification purposes. This technology estimates key body points from video input to compute joint positions, angles, angular velocity, and acceleration. A major advantage of gait recognition is that it operates without physical contact. However, the system remains vulnerable to factors such as different clothing types, camera angles, and partial obstructions.
Technical Features and Accuracy
The developers of this new technology explain:
“The key to solving this challenge lies in extracting stable, individual motion characteristics. To achieve this, we propose a dynamic feature mapping system based on skeletal key points.”By employing dual-stream spatiotemporal convolutional networks, the system integrates features like joint positions, angles, and angular velocities. Attention mechanisms are also applied to adaptively weight the contributions of different body parts.
On the CASIA-B dataset, SKDMap-Net achieved an accuracy of 95.8%, while on the Gait3D dataset, it reached a Rank-1 accuracy of 83.7%. These results mark significant progress in person identification through gait analysis. Interestingly, a research team from Columbia Engineering previously presented data that raised doubts about the reliability of fingerprints for identification. Using AI and a U.S. government database of approximately 60,000 fingerprints, they found that some prints thought to belong to different individuals might actually come from the same person.
In summary, SKDMap-Net opens up new possibilities for person identification with high accuracy, while also highlighting the importance of understanding the limitations of existing methods, such as fingerprinting.
The advancement of identification technologies like SKDMap-Net could have major implications for security and law enforcement. Their contactless nature may reduce risks associated with physical interaction and improve convenience during identification processes. Nevertheless, it is crucial to account for their vulnerabilities, as changes in surveillance conditions can affect result accuracy, questioning their universal applicability across different scenarios.