Research On Detection And Tracking Of Dynamic Objects Based On (VTS) CCTV

Presently, VTS mainly uses radar to detect and track targets in order to achieve the regional monitoring. As CCTV has become an important part of VTS, people can use the valuable achievements of domestic and foreign scholars in the field of video image processing to substitute for radar in achieving target detection and tracking when radar's function is limited.In terms of the above thoughts, the CCTV-based target detection and tracking technology is studied and the following results are achieved in this thesis:Firstly, by analyzing the actual needs, two main methods, background subtraction and frame difference are selected. After analyzing and comparing of the two algorithms, the result is obtained that frame difference which needs fewer operations should be used in rough detection and background subtraction should be adopted when more objects' information is needed. In this section, the result is also obtained that when the background adjusting speed is needed to be refreshed, IIR filter can be used to get background, and contrarily, statistical average is used.Secondly, the method of target feature extraction and target matching between different frames is studied. Based on actual demands, four features (geometrical feature, dymatic feature, statistical feature, moment invariants feature) are extracted .Then a three-level matching principle is established and a relative effective target matching method is obtained.Thirdly , the object tracking algorithms have been studied and the gate algorithm has been chosen as the important one, meanwhile, its key technologies (center obtaining, estimate , gate building) are researched. In conclusion, the thesis showed that Kalman estimate adaptive gate algorithm has a relative high precision which is suitable for complex motion tracking and two-point estimate is useful when the motion is simple. Finally, based on the former principles and algorithms, experiments have been done to detect the dynamic objects in the condition of single object and multiple objects. In terms of tracking algorithm, the dynamic objects in all experiments scenes have been successfully tracked and corresponding tracks are labeled. It is shown that, by experiments, algorithms adopted in the thesis are real-time, robust and effective .