Research On Real-time Detection And Tracking Method Of Satellite Video Dynamic Vehicle And On-orbit Processing

The satellite video coverage area is wide and the time resolution is high.Typical representatives include the Jilin No.1 and Skysat series satellites.As a new type of remote sensing data,it plays an important role in the field of transportation and public security.More complete information of vehicle geographic motion trajectory can be obtained in time by the detection and tracking of vehicle targets in satellite video,which is a new means and development direction for continuous monitoring of ground targets.At present,there are few related researches on real-time detection and tracking methods of satellite video.There are some problems that need to be solved urgently: Satellite video vehicle targets are blurred,its characteristics are not obvious,background illumination changes and parallax effects are serious,resulting in high target miss detection rate and error rate of existing traditional vehicle detection algorithms.Satellite video content is seriously disrupted,and existing video tracking algorithms have difficulty acquiring continuous trajectories of vehicles.There is a bottleneck in the real-time data processing and transmission of satellite video,and a solution for real-time video processing in orbit is needed.In this paper,taking the Jilin No.1 satellite video as an example,the real-time detection and tracking method of satellite video motion vehicles in complex background is studied.The main contributions of the paper include:1)Aiming at the problem of blurred target characteristics of satellite video vehicles,this paper proposes an improved Vibe algorithm combined with the vehicle motion foreground classification method to accurately extract the moving vehicle target by analyzing the shape and motion characteristics of the satellite video moving vehicle,and the experiment proves that the method has good detection rate and accuracy.2)For the problem of serious interference when tracking satellite video moving vehicles,the multi-target continuous tracking of satellite video motion vehicles is transformed into a bipartite graph assignment problem to form a trajectory segment,and the trajectory segment features are used for iterative association.Compared with other tracking algorithms,by using the algorithm proposed in this paper,it is possible to track the geographic trajectory of the target more efficiently and more accurately.3)Aiming at the bottleneck of satellite video data real-time processing and transmission,the prototype of the on-board video processing is designed,and the algorithm is transplanted to the DSP chip.This paper introduces the experiment and result analysis of the video processing prototype.The results show that the multi-core parallel processing achieves the effect of speeding up the operation and reducing the running time.The calculation rate reaches 35 ms per frame,which satisfies the needs of real-time processing in orbit.