Real-Time Pedestrian Detection Based on Dual-Modal Relevant Image Fusion

Fig. 6. Detecting results comparison. (a) GT of VI; (b) VI detection results of baseline; (c) IR detection results of baseline; (d) detection results of baseline after the RAM module is embedded; (e) detection results of the proposed algorithm

Download full size

Table 1. Ablation experimental results of different fusion methods
View table
Table 1. Ablation experimental results of different fusion methods
Fusion method AP₅₀ /% Weight /MB FPS /（frame·s^-1）
Wavelet^［12］ 83.7 / 139
DenseFuse^［14］ 89.7 18.5 126
NestFuse^［23］ 92.1 24.3 98
RFN-nest^［16］ 95.2 33.8 53
Proposed method 94.9 16.8 139

Table 2. Ablation experimental results of different models

View table

Table 2. Ablation experimental results of different models

Method	RAM	Coord-Conv	SGE	Soft-NMS	Input	AP₅₀ /%	Weight /MB	FPS /（frame·s^-1）
YOLOv7-Tiny					VI	48.1	12.3	169
YOLOv7-Tiny					IR	76.3	12.3	169
YOLOv7-Tiny	√				VI+IR	94.9	16.8	139
YOLOv7	√				VI+IR	96.3	48.4	83
YOLOv7-X	√				VI+IR	96.8	82.0	41
Tiny-CoordConv	√	√			VI+IR	95.2	16.8	139
Tiny-SGE	√		√		VI+IR	95.8	16.9	135
Tiny-Soft-NMS	√			√	VI+IR	96.5	16.8	133
Improved model	√	√	√	√	VI+IR	97.3	16.9	124

Tools

Get Citation

Copy Citation Text

Chengcheng Bi, Miaohua Huang, Ruoying Liu, Liangzi Wang. Real-Time Pedestrian Detection Based on Dual-Modal Relevant Image Fusion[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0837014

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites