Athletic performance is affected by a variety of conditions, and effective monitoring, combined analysis, and objective feedback of motor state parameters are important for improving athletes’ athletic skills. The study involved the use of sensing devices to gather data on the movements of athletes during track and field training, calculating the Euclidean distance between their initial positions, and detecting their movement trajectories. During movement, the stride cycle is detected using the time window, and then the movement posture is recognized using the cosine measure method and Softmax regression algorithm. Finally, based on this method, a real-time feedback optimization system for motion data based on intelligent sensing devices has been designed. It is found that the motion detection algorithm proposed in this paper has an accuracy of over 98% in recognizing the motion postures, and can support the motion analysis and feedback of the track and field training process. The mean score of track and field athletes’ performance in motor skills after the application of the real-time feedback optimization system was 85.64 points, which was significantly different from that of athletes with traditional training (P=0.042<0.05). The feedback optimization system for track and field training proposed in this paper can assist coaches in developing scientific and reasonable track and field training methods and promote the innovation and intelligent development of track and field training.