Unsupervised thermal-to-visible domain adaptation method for pedestrian detection

Pedestrian detection is a common task in the research area of video analysis and its results lay foundations wide range applications. It commonly known that under challenging illumination weather conditions, conventional visible cameras perform poorly this limitation could be catered using thermal imagery. But, due to fact annotated datasets are less available compared ones, paper we emphasis need for leveraging information from domain at no additional annotation cost. Precisely, propose adaptation method by incorporating feature distribution alignments into Faster R-CNN architecture different levels two phases network. The resulting proposed adaptive detector has advantage covering aspects shift order improve overall performance. evaluated on KAIST multispectral dataset obtained demonstrate effectiveness improving adaptability domain. Also, means comparisons other existing works, better obtained. Additional experiments conducted further justify results.