OADE-Net: Original and Attention-Guided DenseNet-Based Ensemble Network for Person Re-Identification Using Infrared Light Images

Recently, research on the methods that use images captured during day and night times has been actively conducted in field of person re-identification (ReID). In particular, ReID increasingly performed using infrared (IR) at red-green-blue (RGB) images, addition to ReID, which only uses RGB daytime. However, insufficient IR because their color texture information cannot be identified easily. This study thus proposes an original attention-guided DenseNet-based ensemble network (OADE-Net)—a model can recognize pedestrians times. The OADE-Net consists DenseNets a shallow convolutional neural for (SCE-Net), is used combining two models. Owing lack existing open datasets consist experiments are by creating new dataset retrieved from databases (DBPerson-Recog-DB1 SYSU-MM01). experimental results showed achieved accuracy DBPerson-Recog-DB1 79.71% rank 1, while mean average precision (mAP) 78.17%. Furthermore, 57.30% 1 SYSU-MM01 case, whereas mAP was 41.50%. both higher than score-level fusion state-of-the-art methods.