VSI Pro Report: AI Algorithms for Thermal Image-based Object Detection and Classification

Pro Report 8 Cover - Square.jpg
Pro Report 8 Cover - Square.jpg

VSI Pro Report: AI Algorithms for Thermal Image-based Object Detection and Classification

5,000.00

Thermal cameras (also known as infrared cameras) are generally regarded as one of the essential sensors for pedestrian detection. Thermal cameras are particularly good at differentiating living objects from inanimate objects. The advantage of thermal camera-based sensors over other automotive camera-based sensors is that it provides better visibility in darkness and poor weather conditions. VSI has been examining the use of a thermal camera for object detection and classification by applying machine learning technologies. In this report, VSI discusses the process of creating AI algorithms for thermal image-based object detection, the accuracies of the models we developed, and how we achieved them.

Add To Cart

Table of Contents

  • Introduction

    • Thermal Camera for Automotive

    • Training Dataset for Thermal Camera Images

    • System Prerequisites

  • Procedure

    • Training Dataset Setup: Nvidia DIGITS

    • Training Dataset Setup: Darknet

    • Training Process: Nvidia DIGITS

    • Training Process: Darknet

    • Inferencing Process: Nvidia DIGITS

    • Inferencing Process: Darknet

  • Observations & Results

    • DIGITS Results (Based on Single Classification)

      • Pretrained vs. Not Pretrained

    • Darknet Results

    • Other Observations

      • Image Resizing

      • Anchor Sizes (Darknet Only)

      • DetectNet (DIGITS) vs. YOLOv3 (Darknet)

  • Alternative Solutions

  • Conclusion