The invention provides a new media virtual accessory recommendation system and method based on a liquid crystal dimming film, and the virtual accessory recommendation system is based on the liquid crystal dimming film and the augmented reality technology and is combined with a deep learning algorithm. Compared with subway advertisements in other forms, the virtual accessory recommendation system has better interactivity and customizability and can better attract the attention of users. Meanwhile, the contradiction between the functionality of the metro shielding door and the advertising mediumis solved through the application of the liquid crystal dimming film technology. And the posture information and the clothing attribute information of the user are identified by using a deep learningtechnology, so that different accessories can be recommended to each user, and better customization characteristics are achieved. Through scanning the two-dimensional code in the system, the two-dimensional code jumps to the accessory purchase interface for purchase, and the better interactivity is achieved.