Enhancing the Signal Quality of Low-Visibility Thermal Targets to Improve Recognition and Tracking Performance

Abstract

Thermal imaging cameras capture the infrared radiation emitted or reflected by targets, typically producing grayscale images. While thermal imagery supports night vision and facilitates the detection of living and heat-emitting objects, its application in target detection, tracking, and classification faces significant challenges compared to visible-color cameras. The primary task in target tracking is to extract the target from video frames and continuously obtain its distinctive features, such as location, size, velocity, and orientation.

Conventional tracking algorithms designed for visible images often struggle when applied to thermal images due to the loss of distinctive details, resulting in reduced accuracy and effectiveness. In this study, we propose enhancing weak target features in thermal images by introducing a new attribute called the Effective Thermal Target Center, which is then used to derive the relative orientation of the effective center with respect to the target center. These two features improve the distinguishability of low-feature targets, enabling visible-image-based algorithms to operate effectively on thermal images.

The proposed approach was tested on simulated low-feature thermal targets, where the effective center and target orientation were determined using a simple algorithm. The results were compared with Principal Component Analysis (PCA) for orientation estimation, showing higher orientation detection performance with the proposed method. When integrated with a Multi-Object Tracking (MOT) algorithm supported by a Kalman Filter, the approach demonstrated superior performance in tracking thermal targets.