Evaluation of AI-algorithm for airborne thermal sensor
Publish date: 2022-01-24
Report number: FOI-R--5200--SE
Pages: 36
Written in: Swedish
Keywords:
- RPAS
- recognition
- human performance
- AI-algorithm
Abstract
Air reconnaissance with remote-controlled flight systems is frequently used in military contexts. A long-wave infrared (LWIR) camera similar to that used by the Swedish Armed Forces RPAS system UAV 05B Korpen/RQ-20 Puma was used. In the study presented in this report, ability was added using an automatic measurement recognition algorithm. The ability to recognize targets was tested in two experiments, which was measured in terms of how accurate and how fast vehicles were recognized. The results from experiment 1 show that human performance deteriorates if answers must be given within five seconds, as compared to unlimited time where a more analytical approach can be used. Also, type of vehicle affects performance and lower resolution used to depict vehicles degrades performance. In the experiment 2, data from experiment 1 were used to compare human performance with an algorithm's ability to recognize vehicles. The results show that human performance is better for some vehicles, while the algorithm's performance is better for other vehicles. This means that there is no clear answer as to whether the human or algorithm performs best.