Tracking Technology For Airspace Goals Video

Visual tracking has long been one of the hotspots in domestic and foreign research. It has been widely applied in navigation, security surveillance, medicine and scientific research etc. Especially, the research on video tracking for spatial object has earlier become the key technology in military field. The accuracy and stability of tracking effect depend on the design of algorithms to a great extent. So far, there have been lots of references putting forward optimized algorithm on target detection. However, to improve the tracking performance further will be difficult simply relying on target detection. While organically combining target detection with effective tracking or filter algorithm will probably help to realize stable object tracking. Therefore, aimed at visual tracking for spatial object, this paper will make research on both object detection algorithm and state estimation.Firstly, circling around object detection algorithm, this article designed tracking wave-door and realized gray image binarization by associating the wave-door dimension with the method of sectioning threshold, then the nearest neighbor association algorithm was proposed to identify the true target from those deceptious objects caused by binarization method and the results of emulational experiment demonstrated its validity.Next, in the tracking process, standard Kalman filter algorithm was used firstly to forecast and track the spatial target, then the correlation brought by measure error was analyzed, further continued with the Kalman filter algorithm which considered correlative measure noise was proposed to estimate effectively the motional feature of airspace target and the algorithm not only achieved the goal of accurate tracking but also enhanced the robustness of tracking system.Finally, many different target samples were tracked and tested based on CV-model, Singer model, "Present" model and IMM model so that most suitable target tracking model and self-adaptive filter algorithm for visual tracking could be acquired by analyzing and comparing estimated precision for the same target under each different model.Theoretical analysis and numerical simulation have proved that Kalman filter algorithm considering noise decorrelation had obvious predominance on the forecast errors than traditional Kalman filter algorithm and the accuracy of visual tracking was improved again depend on IMM model. Through this way, this paper may provide useful reference for future visual tracking research.