This paper addresses the bolt assembly task for aircraft panels and proposes a vision-guided composite robotic system, which offers advantages such as a large workspace, accurate target positioning, strong assembly adaptability, and non-contact measurement. To tackle the spatial positioning problem of bolts on aircraft panels, a binocular vision-based method for bolt spatial attitude positioning is proposed. First, an autonomous bolt detection model based on the YOLOv12 neural network is developed, combined with an arc support segment fitting algorithm, to complete the feature extraction of the bolt end-face circle in the image. Then, a matching algorithm based on the SAD-BT cost is proposed to achieve bolt feature matching between the left and right camera images. Finally, using the camera calibration parameters and the triangulation positioning principle, the spatial attitude of the bolt is determined. Experiments are conducted based on assembly accuracy and process requirements, and the results demonstrate that the assembly accuracy is within 1 mm, with an assembly success rate of 100%.