Scholarlink

<p>Perspective-<i>n</i>-Point (PnP) pose estimation is a foundational task in computer vision, supporting augmented reality, visual tracking, and structure-from-motion. Existing linear solvers suffer from rank deficiency in quasi-planar scenes, while global optimal methods incur excessive computational cost. This work presents BAPnP, an efficient and robust linear solver built on barycentric affine invariance. A geometry-guided base selection strategy maximizes the reference basis volume to promote a well-conditioned linear system, and an adaptive reduction handles strictly coplanar cases without singularity. Extensive experiments show that BAPnP retains 100% success rate down to strict coplanarity and runs at <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(4.4\,\mu s\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>4.4</mn> <mspace width="0.166667em" /> <mi>μ</mi> <mi>s</mi> </mrow> </math></EquationSource> </InlineEquation> for <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(N=10\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>N</mi> <mo>=</mo> <mn>10</mn> </mrow> </math></EquationSource> </InlineEquation>, offering a <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(6\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>6</mn> <mo>×</mo> </mrow> </math></EquationSource> </InlineEquation> speedup over SQPnP. The method achieves accuracy comparable to global solvers while retaining the efficiency of linear approaches, making it suitable for real-time AR and visual tracking applications. The source code is publicly available at <a href="https://github.com/lpl8848/BAPnP_Solver">https://github.com/lpl8848/BAPnP_Solver</a>.</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

Bapnp: a barycentric affine invariant linear solver for robust and efficient perspective-n-point pose estimation

  • 期刊论文

  • 发表日期：

    2026年8月3日

查看全文
  • Peilin Luo,
  • Yang Guo

摘要

Perspective-n-Point (PnP) pose estimation is a foundational task in computer vision, supporting augmented reality, visual tracking, and structure-from-motion. Existing linear solvers suffer from rank deficiency in quasi-planar scenes, while global optimal methods incur excessive computational cost. This work presents BAPnP, an efficient and robust linear solver built on barycentric affine invariance. A geometry-guided base selection strategy maximizes the reference basis volume to promote a well-conditioned linear system, and an adaptive reduction handles strictly coplanar cases without singularity. Extensive experiments show that BAPnP retains 100% success rate down to strict coplanarity and runs at \(4.4\,\mu s\) 4.4 μ s for \(N=10\) N = 10 , offering a \(6\times \) 6 × speedup over SQPnP. The method achieves accuracy comparable to global solvers while retaining the efficiency of linear approaches, making it suitable for real-time AR and visual tracking applications. The source code is publicly available at https://github.com/lpl8848/BAPnP_Solver.