LWIR sensor parameters for deep learning object detectors

Abstract

Deep learning has been well studied for its application to image classification, object detection, and other visible spectrum tasks. However, deep learning is only beginning to be considered for applications in the long-wave infrared (LWIR) spectrum. In this work, we attempt to quantify the imaging system parameters required to perform specific deep learning tasks without significant pre-processing of the LWIR images or specialized training. We show the capabilities of uncooled microbolometer sensors for Fast Region-based Convolution Neural Networks (Fast R-CNN) object detectors and the extent to which increased sensitivity and resolution will affect a Fast R-CNN object detector’s performance. These results provide guidelines for design requirements for uncooled microbolometers in industries such as commercial autonomous vehicle navigation that will use deep learning object detectors.

Unconventional Visual Sensors for Autonomous Vehicles

“…Tumas et al [81] present a ZUT dataset containing vehicle odometry and weather measures. Other than re-implementing the CNN detectors for thermal image processing, Grimming et al [82] dive deeply into a thermal camera's physical characteristics, and studied the relation between MTF (modulation transfer function), NETD, and the performance of fast R-CNN object detector [83].…”

Synthetic augmentation methods for object detection in infrared overhead imagery

“…There have been various use cases with infrared imagery in object detection and machine learning in general. [2][3][4][5][6][7][8][9][10] Infrared imagery has been used in object detection, [2][3][4] autonomous driving and vehicle classification, 5,6 and airborne image classification of terrain. 7 Techniques have been introduced to preprocess visible band imagery to use in combination with limited infrared data sets to improve performance.…”