Artificial intelligence for persistent surveillance

The aim of this study the feasibility and the performance of an artificial intelligence-based system that would analyze images acquired every second from a high-flying platform (HALE) with a resolution of about 0.5m. The idea is to first learn the “normal” behavior of targets to obtain or train a model. In a second step, that model is exploited to detect unusual activities, as trajectories start deviating from the expected trajectory. Besides the trajectory, other factors may intervene such as the type of target (pedestrian, car, truck, …), the time of day, the location of the target, …

RMA is in charge of acquiring the image from UAV camera's, performing the extraction and the tracking of targets and finally, developping and assessing the performance of anomaly detection algorithms. Most of these steps will be performed using deep-learning algorithms.

The project DAP/20-02 is financed under the DFR call.