Scholarlink

<p>This paper presents a comparative analysis of various detectors, descriptors, and matching techniques used in computer vision for object tracking in autonomous driving scenarios. The study focuses on classical feature-based methods, including keypoint detectors such as Harris, FAST, and ORB, as well as descriptors like SIFT, SURF, and BRIEF. We also examine feature-matching techniques like brute force and FLANN. Experiments were conducted using the KITTI Vision Benchmark Suite, involving sequential image batches captured from a front-mounted camera in real-world car-following scenarios. The performance of these methods is evaluated based on three main criteria: (1) real-time performance (frame rate and latency), (2) robustness (repeatability, resilience to noise, and false positive/negative rates), and (3) energy efficiency (matches per millisecond, CPU/GPU utilization, and memory footprint). Through extensive benchmarks, we provide insights into the trade-offs between speed, reliability, and resource usage, offering practical guidance for selecting optimal approaches for real-time collision avoidance and other critical autonomous driving tasks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Balancing accuracy and efficiency: detector–descriptor matching strategies for object tracking in autonomous vehicles

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月4日

查看全文
  • Wael A. Farag

摘要

This paper presents a comparative analysis of various detectors, descriptors, and matching techniques used in computer vision for object tracking in autonomous driving scenarios. The study focuses on classical feature-based methods, including keypoint detectors such as Harris, FAST, and ORB, as well as descriptors like SIFT, SURF, and BRIEF. We also examine feature-matching techniques like brute force and FLANN. Experiments were conducted using the KITTI Vision Benchmark Suite, involving sequential image batches captured from a front-mounted camera in real-world car-following scenarios. The performance of these methods is evaluated based on three main criteria: (1) real-time performance (frame rate and latency), (2) robustness (repeatability, resilience to noise, and false positive/negative rates), and (3) energy efficiency (matches per millisecond, CPU/GPU utilization, and memory footprint). Through extensive benchmarks, we provide insights into the trade-offs between speed, reliability, and resource usage, offering practical guidance for selecting optimal approaches for real-time collision avoidance and other critical autonomous driving tasks.