Scholarlink

<p>Multi-object detection and tracking in vehicle footage supports a lot of applications, such as video analytics, smart transportation systems, video surveillance, and robot navigators. Many existing approaches in object tracking and detection result in poor detection performance. To tackle this challenge, this research proposes a Hybrid Fast Regional-You Only Look Once v3 (HFR-YOLO v3) technique for multi-object detection and unscented Kalman filtering-based object tracking in vehicle driving footage. The proposed technique is implemented as follows. The input video is sourced from the dataset and passed through the frame extraction module, which divides the video into individual frames for subsequent processing. Then, the object is segmented from the frame, which is performed by the Entropy Weighting K-means Algorithm. After that, multiple object detection is performed by creating an anchor box using HFR-YOLO v3, which is developed by the amalgamation of YOLO v3 and Fast Regional Convolution Neural Network (FR-CNN). Finally, the multi-object tracking is performed, which is done by Kalman filtering by modifying the moving model. Moreover, the HFR-YOLO v3 accomplished a higher true positive rate (TPR) of 0.977, true negative rate (TNR) of 0.948, and accuracy of 0.958.</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

HFR-YOLO v3 model for multi-object detection and unscented Kalman filtering based object tracking in vehicle driving footage

  • 期刊论文

  • 发表日期：

    2025年1月22日

查看全文
  • Dhanalakshmi Manickaraj,
  • Palvinder Singh Mann

摘要

Multi-object detection and tracking in vehicle footage supports a lot of applications, such as video analytics, smart transportation systems, video surveillance, and robot navigators. Many existing approaches in object tracking and detection result in poor detection performance. To tackle this challenge, this research proposes a Hybrid Fast Regional-You Only Look Once v3 (HFR-YOLO v3) technique for multi-object detection and unscented Kalman filtering-based object tracking in vehicle driving footage. The proposed technique is implemented as follows. The input video is sourced from the dataset and passed through the frame extraction module, which divides the video into individual frames for subsequent processing. Then, the object is segmented from the frame, which is performed by the Entropy Weighting K-means Algorithm. After that, multiple object detection is performed by creating an anchor box using HFR-YOLO v3, which is developed by the amalgamation of YOLO v3 and Fast Regional Convolution Neural Network (FR-CNN). Finally, the multi-object tracking is performed, which is done by Kalman filtering by modifying the moving model. Moreover, the HFR-YOLO v3 accomplished a higher true positive rate (TPR) of 0.977, true negative rate (TNR) of 0.948, and accuracy of 0.958.