Scholarlink

<p>Computer vision algorithms can analyze frames of flooded areas from a UAV video to detect and classify objects of interest. However, using computationally expensive image object detection models in each frame for video scene analysis will significantly impact the speed of execution. This work utilizes the similarity between adjacent frames to reduce computational complexity and improve the speed of the object detector. Two novel algorithms are proposed for estimating pixel displacement in subsequent frames: corner point matching algorithm and the template matching algorithm. They are commonly called pixel displacement algorithms (PDA). This approach allows for the efficient utilization of multiple state-of-the-art image object detection models, as it is independent of the detection model. Improvements in speed of more than 5<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> over continuous frame detection were observed in YOLOv7, with a decay of 2.7% in the AP50 category. Furthermore, the proposed approach resulted in an improvement in accuracy of up to 10% compared to the plain skip approach.</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

Novel Pixel Displacement Prediction Strategies for Fast Object Detection in Flood Video Analysis

  • 期刊论文

  • 发表日期：

    2025年11月29日

查看全文
  • Shubhasree A V,
  • Praveen Sankaran,
  • Raghu C V

摘要

Computer vision algorithms can analyze frames of flooded areas from a UAV video to detect and classify objects of interest. However, using computationally expensive image object detection models in each frame for video scene analysis will significantly impact the speed of execution. This work utilizes the similarity between adjacent frames to reduce computational complexity and improve the speed of the object detector. Two novel algorithms are proposed for estimating pixel displacement in subsequent frames: corner point matching algorithm and the template matching algorithm. They are commonly called pixel displacement algorithms (PDA). This approach allows for the efficient utilization of multiple state-of-the-art image object detection models, as it is independent of the detection model. Improvements in speed of more than 5 \(\times \) × over continuous frame detection were observed in YOLOv7, with a decay of 2.7% in the AP50 category. Furthermore, the proposed approach resulted in an improvement in accuracy of up to 10% compared to the plain skip approach.