Improvised Explosive Devices (IEDs) are one of the primary causes of casualties amongst NATO forces in conflict areas. As IEDs are constructed from a large variety of materials, their type and appearance is a-priori unknown. This makes it difficult to find such objects using standard object-recognition techniques, since it is not possible to 'learn' what an unknown object looks like.
Instead, change detection can be employed to detect IEDs at a safe distance. Here, we assume that the process of hiding an IED causes (small) changes to the environment, such as churned-up earth, newly appeared objects or other disturbances to the environment with respect to a previous patrol. By automatically finding and analyzing such changes, the system is able to warn the driver for potential threats.
Drawing 1: Change detection concept: find changes with respect to a previous patrol.
Change detection is performed by equipping the patrolling vehicle with high-end cameras and an advanced image-analysis system. While driving, the system continuously compares the environment with recordings of the same environment from the past. As soon as a suspicious change is found, the system raises an alarm and shows a visual warning to the operator through an interactive user interface (UI). This interface inside the vehicle shows both the location of the suspicious object, as well as the appearance of the specific location during a previous patrol. Also, by touching the User Interface, the operator canzoom in on any part of the scene, where the UI shows both the present and past situation (Figure 2). This allows the operator to quickly judge if there is a threat, and takeappropriate action.
Drawing 2: Operational concept
Change detection is a challenging process, where the system has to cope with different weather conditions, different driving trajectories and dynamic changes to the scene. Typical examples ofthese challenges are visualized in Figure 3 to 5.
Drawing 3: Different lighting conditions change the visual appearance of the scene, where the change detection system should not be affected by shadows or other lighting effects.
Drawing 4: Different viewpoints change the appearance of the scene, where the relative ordering of objects in the scene may change (2-1-3 vs 1-2-3). This is an example of a strong parallax effect.
Drawing 5: Dynamic changes in the form of new driving tracks.
The prototype system successfully finds test objects as small as 6 x 20 cm at distances of 30 to 40 meter from the vehicle. Furthermore, we achieved:
For security reasons, we do not show detection examples of the full change detection system on this website.