A Systematic Point Cloud Edge Detection Framework for Automatic Aircraft Skin Milling

Zijie Wu; Yaonan Wang; He Xie; Mingtao Feng; Haotian Wu; Chao Ding; Ajmal Mian

cs.CVNov 3, 2026

A Systematic Point Cloud Edge Detection Framework for Automatic Aircraft Skin Milling

Authors: Zijie Wu, Yaonan Wang, He Xie, Mingtao Feng, Haotian Wu, Chao Ding, Ajmal Mian

Abstract

The edge detection technique is an essential step for aircraft skin milling in aviation manufacturing. Most of the current detection methods focus on traditionally defined edge extraction tasks but disregard the crucial systematic requirement of edge milling. In this paper, we proposed a novel edge detection framework for automatic edge milling of aircraft skins. First, an edge probability detector is proposed by the spatial tangent continuity (STC) to provide the essential reference. Second, we propose a hierarchical branch searching (HBS) method to hierarchically strip the desired milling edges from the raw point cloud, which consists of three graded progressive steps: branch backbone generation, branch extension, and branch pruning. We demonstrate the performance of the proposed method on both synthetic models and aircraft skin workpieces. The proposed method outperforms the other baselines and shows accurate edges for the edge milling task.

Figures & tables

Fig. 1: Illustration of the significance of milling edge detection. The point cloud is scanned by the camera and then employed for milling edge detection. The edge information is then applied in robotic edge milling by a machining tool to remove the allowance for final aircraft skin assembly.

Fig. 2: Illustration of the systematic problem formulation: the scanned point cloud from the measured aircraft skin consists of multiple types of points.

Fig. 3: Framework of the proposed milling edge detection.

Fig. 4: Illustration for the definition of spatial tangent continuity.

Fig. 5: Branch Refinement by Graph Optimization. Every row proceeds in parallel, and all linear branches are merged at the end to constitute the final milling edge.

Fig. 7: Precision/Recall evaluation of different edge detectors.

Fig. 8: Error chromatogram of detected edges. The first column shows the sampled point clouds from the three models. The remaining columns show the error chromatogram corresponding to the sampled point clouds for three methods, namely, ED, MS, and HBS. The local details are enlarged on the right of each chromatogram map. These error maps share the same enlarged areas that are indicated with blue arrows.

Model 1
Metrics	Accuracy(%)	Precision(%)	Recall(%)	Error( $mm$ )
ED	98.79	71.36	67.93	0.3088
MS	99.20	86.08	73.53	0.1233
HBS	99.92	98.84	99.17	0.0072
Model 2
ED	98.38	57.88	39.04	0.5972

TABLE I: Quantitative Comparison on edges

Noise ( $\times 10^{-3}mm$ )	2	$\dot{4}$	6	$\dot{8}$	10	$\dot{12}$	14	$\dot{16}$	18	$\dot{20}$
Error ( $mm$ )	0.0093	0.0094	0.0096	0.0101	0.0104	0.0108	0.0117	0.0121	0.0126	0.0130
Outlier ( $\times\%$ #pts)	2	$\dot{4}$	6	$\dot{8}$	10	$\dot{12}$	14	$\dot{16}$	18	$\dot{20}$
Error ( $mm$ )	0.0092	0.0093	0.0092	0.0092	0.0092	0.0094	0.0095	0.0094	0.0095	0.0093

TABLE II: Noise Resistance with irregularly sampled

PD	CV	GR	Accuracy(%)	Precision(%)	Recall(%)	Error( $mm$ )
$\checkmark$	$-$	$-$	99.83	96.22	96.65	0.0214
$\checkmark$	$\checkmark$	$-$	99.87	96.57	98.10	0.0187
$\checkmark$	$-$	$\checkmark$	99.84	98.31	97.13	0.0113
$\checkmark$	$\checkmark$	$\checkmark$	99.88	98.51	98.18	0.0092

TABLE III: Ablation study for different module combinations

Free parameters	Value	Precision(%)	Recall(%)	Time(s)
$E_{min},E_{max}$	0.8,0.9	92.87	93.56	0.46
	0.6,0.9	96.16	94.78	0.59
	0.6,0.8	98.49	98.17	0.72
	0.4,0.8	98.51	98.18	0.82
	0.2,0.8	98.51	98.18	3.58
	0.4,0.6	98.51	98.18	1.24

TABLE IV: Impact of hyper-parameters

Point Number(#pts)	Running Time(s)	Error(mm)
$1\times 10^{4}$	0.13	0.0209
$5\times 10^{4}$	0.22	0.0099
$10\times 10^{4}$	0.28	0.0093
$50\times 10^{4}$	0.51	0.0092
$100\times 10^{4}$	0.82	0.0092

TABLE V: Time cost of our HBS method to different point numbers

Fig. 10: Edge Detection using HBS on Skin Models.

Circle	$Error$ ( $mm$ )	$d_{z}$ ( $mm$ )	$Std$ ( $mm$ )	S.t.
EV	0.1124	0.0678	0.1059	23 times
MS	0.4912	2.7123	0.4219	11 times
HBS	0.0112	0.0096	0.0173	11 times
Square	$Error$ ( $mm$ )	$d_{z}$ ( $mm$ )	$Std$ ( $mm$ )	S.t.
EV	0.2015	0.0893	0.2876	25 times
MS	0.5789	3.2310	0.5134	13 times

TABLE VI: Performance comparison on skins

Fig. 11: Edge detection in diversified scenarios using HBS.

Metric	$Error$ ( $mm$ )	$Std$ ( $mm$ )	Flatness $d_{z}$ ( $mm$ )
Profile inspection	0.0013	0.0016	0.0009
Metric	$Error$ ( $mm$ )	$Std$ ( $mm$ )	GT/Avg.Diameter( $mm$ )
Hole diameter measurement	0.0016	0.0007	1.5000/ 1.5014
Metric	$Error$ ( $mm$ )	$Std$ ( $mm$ )	1-NN Gap( $mm$ )
Assembly gap detection	0.0011	0.0014	4.0010

TABLE VII: Three Extended Applications using HBS

Edge Detection

See all 13 papers

EdgeFormer: local patch-based edge detection transformer on point clouds
Apr 23, 2026Yifei Xie, Zhikun Tu, Tong Yang +23D ViTs Point Cloud Learning
School of Information Science and Technology, Northwest University, 2026 Xuefu Street, Xi’an, 710127, Shannxi, China.
TriCLE: Tri-Modal Vision-Language Reasoning for Edge-Deployed Fine-Grained Clustering
Aug 4, 2026Kishor Datta Gupta, Md. Mahfuzur Rahman, Fahad Rahman +4Efficient VLM Inference Clustering
Department of Cyber Physical Systems Clark Atlanta University Atlanta, GA, USA · Department of Computer Science and Engineering United International University Dhaka, Bangladesh · Department of Electrical and Computer Engineering North South University Dhaka, Bangladesh
MatchED: Crisp Edge Detection Using End-to-End, Matching-based Supervision
Feb 24, 2026Bedrettin Cetinkaya, Sinan Kalkan, Emre AkbasEdge Detection Edge Inference
Dept. of Computer Engineering and METU ROMER Robotics Center, Ankara, Turkey

Apr 23, 2026J/K move · Enter open · S save

EdgeFormer: local patch-based edge detection transformer on point clouds

Yifei Xie, Zhikun Tu, Tong Yang, Yuhe Zhang+1

School of Information Science and Technology, Northwest University, 2026 Xuefu Street, Xi’an, 710127, Shannxi, China.

3D ViTs Point Cloud Learning Point Cloud Understanding

Open paper AskPDF arXiv

Edge points on 3D point clouds can clearly convey 3D geometry and surface characteristics, therefore, edge detection is widely used in many vision applications with high industrial and commercial demands. However, the fine-grained edge features are difficult to detect effectively as they are generally densely distributed or exhibit small-scale surface gradients. To address this issue, we present a learning-based edge detection network, named EdgeFormer, which mainly consists of two stages. Based on the observation that spatially neighboring points tend to exhibit high correlation, forming the local underlying surface, we convert the edge detection of the entire point cloud into a point classification based on local patches. Therefore, in the first stage, we construct local patch feature descriptors that describe the local neighborhood around each point. In the second stage, we classify each point by analyzing the local patch feature descriptors generated in the first stage. Due to the conversion of the point cloud into local patches, the proposed method can effectively extract the finer details. The experimental results show that our model demonstrates competitive performance compared to six baselines.

Abstract

Edge Detection

Explore similar work

EdgeFormer: local patch-based edge detection transformer on point clouds

TriCLE: Tri-Modal Vision-Language Reasoning for Edge-Deployed Fine-Grained Clustering

MatchED: Crisp Edge Detection Using End-to-End, Matching-based Supervision