Research And Implementation Of Real-time Object Tracking System Based On Self-Balanced Gimbal

With the rapid development of computer technology,computer vision also enters new development level.Mature algorithms(such as Face recognition,Traffic detection)are coming to our life gradually,and with the popularity of Deep Learning these years,intelligent monitoring system will have a huge future based on powerful internet.This research is about real-time object tracking based on self-balanced gimbal,differing from traditional tracking system,this system can avoid motion blur of image,whose camera platform is self-balanced,improved the quality of the image,and reduced the disturb for tracking algorithm as well.This research cut into two parts: self-balanced gimbal and real-time object tracking.The first part includes brushless DC motor control,attitude fusion,and decoupling control of the three axis gimbal.The second part is about the tracking framework includes feature point detection and motion estimation,through these two modules,two object results can be get,then the system picks the best result as the true result based on Bhattacharyya coefficient calculation.On the basis of related theory research,the whole system is implemented,including software and hardware,and some necessary tests have done: camera platform is of good performance to stabilize the image,the maximum error of self-balanced angle of each axis is no more than ±20.We also tested the tracking algorithm,whose processing speed reaches12 FPS or more.This algorithm can hold scale change and rotation in plane,and also has illumination and occlusion tolerance.Whafs more,the recognition rate with self-balanced gimbal is higher than situations without one according to the test.Above all,our system is of great vibration tolerance and tracking performance,and can be applied to some complex monitoring occasions.