Development of Robust Vehicle Tracking Algorithms in Aerial Image Sequences

Kurzfassung

The thesis aims at working out feasible, rational and robust continuously vehicletracking algorithms with aerial images. It concentrates mainly on the travel time of a vehicle travelling in occluded area, such as tunnels and roads under bridges. At first traffic states are categorised with non-impeded or congested states. The classification of traffic condition depends on the traffic states at entrances and exits of occluded area. Then velocity of the tracked vehicle will be derived from the densityflow relation illustrated in the fundamental diagram. Here three approaches are discussed to predict the travel time in occluded area: the statistical analysis, the cell transmission model and the microsimulation. The statistical analysis uses a velocity, which is generated stochastically with means and standard deviations of the observed velocity. The cell transmission model is applied for the prognosis of changes of density and speed on road sections (cells). In the microscopic traffic simulation, flows and traffic control methods are defined and calibrated according to preprocessed data from aerial images, so that the travel time exported from the simulation is regarded as the time to guide a glider for continuous tracking. In the case of congestion, the propagation of the congestion is assessed according to both the shock wave theories and the receiving function in the cell-transmission model (CTM). With the consequence of the predictive approaches, the travel time is compared and assessed by the detecting and tracking data from aerial images. These images are taken in Cologne and Munich by 3K cameras on a glider of DLR. At the end, the deflection of the predicting result caused by false positives and negatives in detecting and tracking is analyzed to argue feasibility of the approach with CTM model, while available databases for networks and traffic-control methods decide the possibility of the use of the microsimulation for tracking vehicle continuously.