Real-time object recognition and localization in deep-sea underwater grasping operations using binocular stereo vision are challenging tasks. Owing to the unknown nature of objects prior to deep-sea underwater operations, it is generally challenging to establish object image datasets in advance. This study adopted a scheme combining an object tracking algorithm and a feature extraction algorithm to identify deep-sea underwater operation objects in real-time accurately. The object tracking algorithm identifies the object in real-time to determine its position and scale in the image. The feature extraction algorithm extracts features from the key points of the object, performs feature matching and screening processes on the left and right image to realize object localization, and calculates the object's three-dimensional coordinate by matching the key points of the left and right image patches as homologous points. Experiments conducted in a laboratory pool environment showed that the method proposed in this study could identify and locate typical grasping operation objects in real-time and at the millimeter level, complete the recognition and location of the typical object in real-time, and guide the underwater manipulator during grasping operations.