An MoD/DSTL funded project through the Autonomous Systems Underpinning Research (ASUR) programme
Small unmanned air systems (SUAS) cannot perform computationally intensive tasks on board because of the physical constraints imposed upon them (weight, power consumption, responsiveness). These systems may only be capable of transmitting raw data to ground terminal systems, often intermittently when the quality of the communication link is compromised. Consequently CPU intensive tasks are offloaded to ground-based resources, often under supervision of a human expert. Extraction of vehicle movement data from videos and detection of anomalous vehicle tracks are computationally demanding tasks. The proposed project envisages the representation of observed vehicle trajectories as graphs to be analysed with novel graph mining approaches in order to detect both structural as well as numerical discrepancies in vehicle movements. The programme will identify the most effective way of transforming vehicle tracks into directed graphs with numerical labels and evaluate different graph-mining approaches. In Phase I of the project the system will be trained and tested using synthetic data generated using a traffic-flow simulator.
Personnel Involved (QUB)