Programmable metasurfaces allow real-time electromagnetic (EM) manipulation in a digital manner, showing great potential to construct advanced multifunctional EM devices. However, the current programmable metasurfaces typically need human participation to achieve various EM functions. In this Letter, we propose, design, and construct a self-adaptive metasurface platform that can achieve beam control automatically based on image recognition. Such a platform is composed of a metasurface with 36×36 active units, a single-board computer, and two independent cameras that can detect the position of the objects. The single-board computer, Raspberry Pi, is used to calculate the information of the objects and generate the coding sequences to control the digital metasurface based on a low complexity binocular localization algorithm. Such a smart metasurface platform can generate different beams according to the location of the receiver and can be used as intelligent antennas in future communications and radars.