Noise is crucial for aircraft, and the flyover noise test is used to measure the distribution of aircraft noise. European and American countries have extensively conducted leap tests, and almost all emerging large aircraft have undergone leap tests, including the C919. During the leap test, it is necessary to perform three-dimensional positioning of the aircraft and a set of independent positioning schemes that do not rely on the aircraft’s own GPS system to ensure the accuracy and synchronization of the test data. The existing computer vision based solutions do not meet the positioning accuracy requirements in leap noise testing due to inaccurate steps in aircraft target detection and image localization. This article improves this step based on the most widely used YOLO model and develops a computer vision based aircraft 3D positioning method. Firstly, a proprietary dataset was constructed and expanded using low-cost methods for the three-dimensional positioning problem of aircraft; Next, a proprietary dataset was used to train the YOLO model, and the trained YOLO model was used for aircraft keypoint detection; Then, based on instance segmentation, key points were optimized to achieve accurate aircraft target detection and image localization. Finally, the effectiveness of the entire method was verified in dual camera and model flight experiments. Under the condition of a measurement scale of 1:100, the three-dimensional positioning error of the aircraft can be controlled within 1.18 m, achieving high-precision continuous positioning without the need for satellite signals.