This invention discloses a
train driving action training
system, including an
image acquisition module that constructs a dedicated dataset for standardized operational gestures specific to locomotive
crew members, retaining only the driver's hands, arms, and key gesture areas to eliminate irrelevant background interference; an image
annotation module for annotating, quality checking, and correcting the original gesture images; and a model training module that employs the YOLO11s
object detection algorithm, enhancing operation by disabling redundant data to ensure the accuracy of
gesture recognition. Through the collaborative work of the
image acquisition module, model training module, and real-time recognition training module, and by employing the YOLO11s
object detection algorithm and optimizing training for locomotive driving-specific gestures, this invention achieves automated, batch, real-time, and
continuous monitoring of locomotive driver finger and verbal gestures. It effectively solves the technical pain points of low efficiency and insufficient accuracy in traditional manual
verification, significantly improving the efficiency and accuracy of locomotive
crew member gesture checks.