Design of an Adaptive Model for Intelligent Car Seats Based on Human Pose Recognition

Electric vehicles (EVs) are moving toward greater intelligence and personalization, but the dynamic adaptability of current vehicle seats no longer meets drivers' comfort and safety needs. To address insufficient posture adaptation and low recognition accuracy for occluded joints, this study proposes an intelligent seat adaptive adjustment model on the basis of human pose recognition. Kinect V2 was used to collect drivers' posture data, and features were extracted through a cascaded pose analysis network (PSN) integrating a spatial transformer network, squeeze-and-excitation module, and convolutional block attention module. The network recognized 16 joint points with an average accuracy of 0.87, a PCK@0.2 (percentage of correct keypoints within 20% of limb length) of 0.92, and a mean per joint position error (MPJPE) of 3.2 mm. Considering biomechanical and vehicle physical constraints, a seat adjustment model was established and optimized using the analytic hierarchy process (AHP) to coordinate seat fore-aft position, height, and backrest angle. Simulated driving experiments suggested preliminary effectiveness: Subjective comfort increased by about 30%, lumbar muscle activity decreased by 39.5%, and posture estimation plus parameter generation remained within 0.3 s. These findings indicate preliminary feasibility for intelligent seat posture assistance under simulated conditions, whereas practical safety validation in real driving conditions remains for future work.