| 摘要: |
| 火箭发动机喷管是控制火箭飞行姿态的主要部件，因此地面测试中的喷管极性测试是一个重要步骤。当前的发动机喷管极性测试依靠人工观测进行判断，对于摆动快、摆角小的一些动作难以清晰辨别。基于工业摄像头系统，利用计算机视觉技术对每一级喷管的运动极性进行实时地自动化识别。在算法设计中，利用基于YOLOv3-tiny的目标识别技术与基于Farneback光流法的运动检测技术，有效地判断出每一个喷管的运动轨迹与极性。同时为了方便地面测试人员使用，将算法、摄像头控制等集成于软件平台，形成软硬件相互协同的一体化系统，做到极性测试自动化、可记录、可追溯。经测试,提出的方法极性识别准确率达到100%，8路摄像头同时工作的视频流识别效率达到20 帧/s以上。表明该系统能够提高极性测试效率，保证测试可靠性，为火箭型号测试无人值守提供了解决方案。 |
| 关键词: 目标识别 运动检测 自动化测试 火箭喷管 极性测试 |
| DOI： |
|
| 基金项目:民用航天“十三五”技术预先研究项目 |
|
| Research on Automatic Recognition Method of Rocket Nozzle Polarity Based on Computer Vision |
| FANG Yiwei,WANG Guan,YI Hang,ZHANG Heng |
| (Shanghai Jiao Tong University;Beijing Institute of Astronautical Systems Engineering) |
| Abstract: |
| The rocket launch engine nozzle is a main component that controls the rocket's flight attitude, so the nozzle polarity testing is an important step in rocket ground test. Currently, engine nozzle polarity testing relies on manual observation for judgment, and it is difficult to clearly distinguish some movements with fast swing rate and small swing angle. This research is based on an industrial camera system, using computer visual technology to automatically recognize the movement polarity of each stage nozzles in real time. In the algorithm design, the target recognition technology based on YOLOv3-tiny and the motion detection technology based on the Farneback optical flow method are used to effectively determine the trajectory and polarity of each nozzle. In order to facilitate the usage of ground test operators, the algorithm and the camera control are integrated into the software platform, forming an integrated system of software and hardware coordination. Based on this, the polarity test can be recordable and traceable. According to the test, the polarity recognition accuracy of the proposed method reaches 100%, and the video stream recognition efficiency of 8 cameras working at the same time reaches more than 20 fps. The system is able to improve test efficiency as well as the measurement reliability，and provide a solution for unattended rocket test. |
| Key words: Target recognition Motion detection Automated test Rocket nozzle Polarity test |