Gait is an important indicator that reflects an individual’s physical or mental health status. This study employed smartphones to measure participants’ gait across various emotional states to develop gait-based emotion recognition. We also assessed emotion recognition based on real-world gait and the impact of emotion feedback to validate long-term emotion motoring. Results indicated that when the emotion deviated from the neutral state, the gait parameters also deviated and the manifestation in gait was related to both the valence and arousal of the emotion. Among positive emotions, high and low arousal emotions showed significant differences in gait parameters. No significant changes were observed in gait parameters after emotional feedback. The recognition accuracy of the gait-based emotion recognition model was higher for high-arousal emotions, but not sufficient for low-arousal emotions. Based on the dimensional theory of emotion, this study focused on the healthy adult population, established a gait-based emotion prediction model, and further validated it in daily life.