An Efficient Object Detection and Classification from Restored Thermal Images based on Mask RCNN

• Published in: International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (Online) pp. 639 - 645
• Main Authors: Thenmozhi, E., Karunakaran, A, Arunkumar, J.R., Chinnammal, V., Kalaivanan, C., Anitha, G
• Format: Conference Proceeding
• Language: English
• Published: IEEE 10.11.2022
• Subjects: Cameras; deep learning; Generative adversarial networks; Image deblurring; Soil; subject and thermal reflection detection; Superresolution; Surface reconstruction; Surveillance; Thermal analysis; Thermal images
• ISSN: 2768-0673
• DOI: 10.1109/I-SMAC55078.2022.9987422

In recent years, thermal cameras are extensively employed in several industries, including biometrics, intelligent surveillance, and health monitoring. The thermal cameras' exorbitant price, meanwhile, makes them difficult to obtain. Furthermore, blurring brought on by camera movement, object movement, and focus settings is a problem with thermal photos. There haven't been many research on thermal image-centered picture restoration that focus on such issues. Additionally, it is critical to accelerate the processing capability of image treatment technologies in order to work in tandem with techniques like object tracking and activity detection that make use of temporal data from thermal recordings. Furthermore, no research has been done on the use of thermal pictures for super-resolution rebuilding and deblurring. Due to the inability to discern reflected on the soil surface or wall caused by the heat emitted by the item, previous research on object recognition using thermal imaging include inaccuracies. This paper suggests a deep learning-based technique for thermal image reconstruction that combines deblurring and super-resolution reconstruction in one step. Recent advances in deep learning have shown that approaches based on generative adversarial networks (GANs) perform well in image-to-image translation challenges because they can maintain texture features in pictures and produce finer, more convincing textures than traditional feed forward encoders. This research study suggests a deblur-SRRGAN for thermal image restoration in light of the benefits of GAN. Additionally, a MR-CNN is also recommended to perform object recognition in the thermal image reconstruction.