Open Access

Optimisation of highway vehicle occlusion recognition based on attention and multitasking approach

Shifeng Feng

Shifeng Feng

Hebei Expressway Group Limited Qingyin BranchShijiazhuang, China
Search for this author on
Sciendo|Google Scholar
Feng, Shifeng
  
Mar 17, 2025
Optimisation of highway vehicle occlusion recognition based on attention and multitasking approach's Cover Image
About this article
Previous Article
Next Article
Cite
Download Cover
Published Online: Mar 17, 2025
Received: Oct 18, 2024
Accepted: Feb 12, 2025
DOI: https://doi.org/10.2478/amns-2025-0180
Keywords
© 2025 Shifeng Feng, published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1.

Key point extraction and coordinate positioning of vehicles
Key point extraction and coordinate positioning of vehicles

Figure 2.

The structure of CAMTS
The structure of CAMTS

Figure 3.

Structure of two line multi-task vehicle recognition model
Structure of two line multi-task vehicle recognition model

Figure 4.

The structure of MAFF
The structure of MAFF

Figure 5.

A two-line multitasking vehicle occlusion recognition modeling process
A two-line multitasking vehicle occlusion recognition modeling process

Figure 6.

Ablation test results
Ablation test results

Figure 7.

PR curves and area test results for each module
PR curves and area test results for each module

Figure 8.

The occlusion detection results of different detection methods
The occlusion detection results of different detection methods

Comparison of different methods for detecting occluded vehicles

Data set Model P/% R/% F1/% MSE MAE
COCO Mask R-CNN 79.33 82.29 80.81 0.04 0.06
Faster R-CNN 84.69 84.17 84.43 0.02 0.04
YOLOv7 89.75 88.74 89.32 0.02 0.02
Our method 93.28 92.47 92.82 0.01 0.02
ApolloScape Mask R-CNN 84.25 86.54 85.41 0.03 0.06
Faster R-CNN 86.69 88.15 87.36 0.03 0.05
YOLOv7 89.51 90.33 89.51 0.03 0.04
Our method 92.23 93.82 92.62 0.01 0.02

Comparison of model detection results under different occlusion conditions

Occlusion rate Method Color precision/% Vehicle type precision/% Detection time/s
20% Mask R-CNN 79.08 87.22 1.04
Faster R-CNN 76.33 72.23 1.46
YOLOv7 82.81 73.52 1.22
Our method 91.74 93.49 0.29
40% Mask R-CNN 80.13 79.45 0.93
Faster R-CNN 75.62 86.51 0.77
YOLOv7 82.77 89.52 0.61
Our method 90.06 89.06 0.44
60% Mask R-CNN 80.45 80.94 0.55
Faster R-CNN 83.92 78.25 1.94
YOLOv7 76.76 83.49 0.74
Our method 94.56 88.67 0.21
80% Mask R-CNN 78.77 81.47 1.28
Faster R-CNN 82.36 84.22 1.18
YOLOv7 84.58 87.11 1.92
Our method 92.52 88.16 0.53
Language:
English
Publication timeframe:
1 times per year
Journal Subjects:
Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other
Journal RSS Feed