Abstract: Implementations of the present application provide a solution for obtaining a stable frame. In the solution, a video stream is obtained, from which a first and second stable frame are determined using at least position-based offset values between one or more frames. The first and second stable frames correspond to a first time interval and second time interval, respectively. The first time interval corresponds to a time before a first vibration is generated by opening a door of a visual vending machine. The second time interval corresponds to a time after the first vibration and before a second vibration generated by closing the door. The two stable frames can be used for image recognition within the visual vending machine to improve accuracy of product recognition.