Depth estimation of thermal infrared images based on self-supervised learning

Fig. 6. Input image and distance estimation results，（a） the input image and the ground truth，（b） the result of distance estimation by the proposed method，（c） the result of distance estimation by HR-Depth，（d） the result of distance estimation by Monodepth2

Download full size

View in Article

Table 1. Related parameters of thermal infrared cameras

View table

View in Article

Table 1. Related parameters of thermal infrared cameras

参数	FLIR-Tau2	FLIR-A35
图像分辨率	640×512	320×256
相机参数	HFOV 45° VFOV 37° 13 mm f/1.0	HFOV 48° VFOV 39° 9 mm f/1.0
相机内参数矩阵	$[\begin{matrix} 0.669 & 0 & 0.5 & 0 \\ 0 & 0.828 & 0.5 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{matrix}]$	$[\begin{matrix} 0.6403 & 0 & 0.5 & 0 \\ 0 & 0.8003 & 0.5 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{matrix}]$
图像采样率	30 Hz	30 Hz

Table 2. Training parameters
View table
View in Article
Table 2. Training parameters
参数数值
ResNet层数 18
学习率 0.000 1
迭代次数 20

Table 3. Error Rates of depth estimation for different networks
View table
View in Article
Table 3. Error Rates of depth estimation for different networks
方法 Proposed HR-Depth monodepth2
E 19.58% 20.09% 21.68%

Table 4. Proportions of different network error distribution intervals （%）
View table
View in Article
Table 4. Proportions of different network error distribution intervals （%）
方法 E
<10% <20% <30% >30%
Proposed 41.67% 66.67% 90.00% 10.00%
HR-Depth 36.67% 63.33% 86.67% 13.33%
monodepth2 25.00% 58.33% 85.00% 15.00%

Tools

Get Citation

Copy Citation Text

Meng DING, Song GUAN, Shuai LI, Kuai-Kuai YU, Yi-Ming XU. Depth estimation of thermal infrared images based on self-supervised learning[J]. Journal of Infrared and Millimeter Waves, 2023, 42(6): 906

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites