All-Weather Pedestrian Detection Based on Double-Stream Multispectral Network

Recently, advanced driver assistance systems (ADAS) have attracted wide attention in pedestrian detection for using the multi-spectrum generated by multi-sensors. However, it is quite challenging image-based sensors to perform their tasks due instabilities such as light changes, object shading, or weather conditions. Considering all above, based on different spectral information of RGB and thermal images, this study proposed a deep learning (DL) framework improve problem confusing sources extract highly differentiated multimodal features through multispectral fusion. Pedestrian methods, including double-stream network (DSMN), were used fusion detector with Yolo-based (MFDs-Yolo) information. Moreover, self-adaptive weight adjustment method improved illumination–aware (i-IAN) later strategy, making modes complimentary. According experimental results, good performance was demonstrated public dataset KAIST FLIR, even performed better than most miss rate (MR) ([email protected]) evaluation system.