Crowd Counting Study Based on Low Light Image Enhancement
Qiao Hu, Guangping Li · 2023
In order to improve the uniformity of data brightness, we propose a crowd counting architecture based on low-light image enhancement in this research. Mainly, a module with curve estimation is added to the architecture, the RGB and T images in the RGBT-CC dataset are enhanced at the light level. The effective channel attention mechanism and exposure control loss are added to suppress low-light/dark areas while reducing background noise. Based on the BL as the backbone, a lot of experiments are conducted. The results show that the data enhancement at the light level of the dataset can improve the illumination uniformity and reduce crowd counting error, proving that the model has higher accuracy and robustness.