Re-identification of previously observed targets in EO/IR-data from UAV. A Survey

The re-identification method should not rely on the fact that the vehicle is stored in a target database prior to the mission start. The idea is that a UAV that is tracking a target should be able to perform another task and then return, detect vehicles, and among them re-identify the previously tracked vehicle. Thereafter the UAV can continue with the tracking of the vehicle. It is the re-identification step in this scenario that has been the topic of this report. To be able to re-identify the target we must collect target features during the continuous tracking. For a more robust re-identification new target features should be assembled when they appear. A good re-identification method should be able to handle changes in pose, scale, lighting conditions and so on. We have presented some approaches for object recognition and investigated a couple of re-identification methods of vehicles from UAV that was developed in the VIVID program. In the future we will implement a re-identification algorithm that uses one of the VIVID methods as a model but also incorporates other recognition steps and methods.