Cross-scale infrared pedestrian detection based on dynamic feature optimization mechanism

Input：红外图像 $I_{i r}$ ，最大优化迭代次数 $M a x_i t e r a t i o n$ ，寻优参数 $α$ ，搜索种群 $X_{A t t a c k e r}$ ， $X_{C h a s e r}$ ， $X_{B a r r i e r}$ 和 $X_{D r i v e r}$ .

1. LPM模块

构造四叉树-贝塞尔插值算子重构初始红外背景图像；

引入引导滤波平滑噪音并得到红外亮度特征图像 $I_{L i r}$ 和红外背景图像 $I_{B i r}$ ；

2. EG-Chimp优化模型

构建动态特征优化图像： $I_{O i r} = α \times I_{L i r} + I_{B i r}$ ；

设计目标函数对参数 $α$ 寻优： $F = m i n {L_{S F} + λ L_{C O N}}$ ；

While $t < M a x_i t e r a t i o n$

For each chimp

计算各人猿种群的位置向量；

更新 $f, m, c, a, D$ ；

End For

For each search chimp

更新目前搜索种群的位置向量；

End For

更新 $X_{A t t a c k e r}$ ， $X_{C h a s e r}$ ， $X_{B a r r i e r}$ 和 $X_{D r i v e r}$ ；

$t = t + 1$ ；

End While

Output：动态特征优化图像 $I_{O i r}$

Table 1. Software and hardware platform configuration
View table
View in Article
Table 1. Software and hardware platform configuration
Configuration Version parameter
Operating system Microsoft Windows 10
GPU NVIDIA GeForce GTX 1660 Ti
CPU Intel Core i5-10400F@2.90 GHz×6 CPUs
CUDA 11.1
Deep learning framework Pytorch

Table 2. Average values of evaluation indexes for 1 000 images
View table
View in Article
Table 2. Average values of evaluation indexes for 1 000 images
Image Entropy Brenner DCT Variance
IR 5.65 4.72×10⁶ 4.69×10³ 1.06×10⁸
Optimized IR 6.04 2.84×10⁷ 4.87×10³ 1.92×10⁸

Table 3. Improved module validation
View table
View in Article
Table 3. Improved module validation
Component Method
GIOU √ √ √ √
CIOU √ √ √ √
DFOM √ √ √ √
YOLOv5 √ √ √ √
CSFF-BiFPN √ √ √ √
mAP@.5/% 88.3 88.8 89.4 89.1 90.2 89.8 89.5 90.7

Table 4. Comparison results of different detection algorithms

View table

View in Article

Table 4. Comparison results of different detection algorithms

Method	Backbone	mAP@.5/%	mAP@.5：.95/%	Time/s
Faster-RCNN	ResNet50	72.9	—	0.163
SSD	VGG	75.1	—	0.087
RetinaNet	ResNet50	74.8	36.3	0.140
Sparse R-CNN	ResNet50	84.3	42.9	0.148
VarifocalNet	ResNet50	86.2	45.3	0.172
TOOD	ResNet50	88.9	46.1	0.156
YOLOv4-CLAHE	CSPDarkNet	87.9	45.9	0.042
I-YOLO	DarkNet	88.9	45.8	0.024
TC-Det	DarkNet	89.2	46.2	0.149
CSNet（Ours）	Focus+ CSPDarkNet	89.5	46.2	0.011
DFOM-CSNet（Ours）	Focus+ CSPDarkNet	90.7	46.8	0.013

Tools

Get Citation

Copy Citation Text

Shuai HAO, Tian HE, Xu MA, Lei YANG, Siya SUN. Cross-scale infrared pedestrian detection based on dynamic feature optimization mechanism[J]. Optics and Precision Engineering, 2022, 30(19): 2390

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites