Research‎ > ‎Defense Imaging‎ > ‎

Thermal Infrared Object Tracking

IR 물체 추적은 다양한 적외선 센서를 통해 촬영된 영상을 분석하여 목표 물체를 추적하는 분야입니다. 전투기 및 미사일 등의 전술 무기는 수 백도의 고온을 유지하기 때문에 가시광선을 비롯한 다른 영역대의 영상에 비해 적외선 영상에서 잘 드러나므로 적외선 영상 분석 시스템은 군사 영상 분석에 적합합니다. 이와 관련하여 가시광선 영역의 영상을 적외선 영상으로 변환하는 방법(Image translation)과 이를 활용한 딥러닝 기반 물체 추적 기술 연구를 진행하고 있습니다.

  • RGB image to thermal infrared (IR) image translation (in progress now)







  • Thermal infrared object tracking dataset (VOT-TIR) [Reference]

The VOT-TIR dataset consists of 20 sequences of which eight has been recorded specifically for this dataset. The other twelve sequences have been collected from different sources including Termisk Systemteknik AB, the Department of Electrical Engineering at Linköping University, the School of Mechanical Engineering at the University of Birmingham, ETH Zürich, Fraunhofer IOSB, Aalborg University, and finally the EU FP7 project P5.

The raw signal values from a thermal infrared sensor are typically stored in 16-bit format. Since not all trackers can handle 16-bit data and for the purpose of visualization, all sequences in the dataset have been truncated to 8-bit. In practice, this is a common procedure since not all sensors give access to the 16-bit values. Therefore, the sequences are not radiometric (the corresponding temperature value is unknown) and the dynamic may adaptively change during the course of a sequence.

The sequences of VOT2016 dataset are the same sequences of VOT2015 dataset. However, the GT of VOT2016 is more accurate than the GT of VOT2015 dataset which has an impact on the evaluation. The VOT-TIR2016 dataset was updated with new sequences.

    • Sample images






  • Object tracking results