Enhanced Thermal Object Detection and Classification with MobileNetV3: A Cutting-Edge Deep Learning Solution
Ravina Gupta, Sarika Jain, Manoj Kumar · 2024
Object detection and classification have become significant areas of interest in the domains of image processing and computer vision. Their widespread adoption extends to various application domains, including face detection, autonomous vehicles, pedestrian detection, and security surveillance systems. Conventional methods such as support vector machines (SVM), Gaussian mixture models (GMM), and background subtraction have been around for a while. However, they have limitations such as object overlap and are prone to distortion caused by ambient factors including smoke, fog, and changing illumination conditions. In this research, the performance of a pre-trained deep learning model MobilNetV3 was evaluated on ground-based thermal images consisting of persons and cars through a thermal camera. The detection is not affected by smoke and poor weather conditions. The optimized thermal imaging dataset’s test and validation data were used to assess the model’s performance. Our results demonstrate consistent accuracy for both person and car classes. Notably, the model exhibits superior performance in classifying the person class, achieving a higher recall value of $\mathbf{9 8. 6 \%}$.