http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
RGB 및 열화상 이미지 퓨전을 활용한 야간 보행자 검출 네트워크
장준보(Junbo Jang),김희광(Heegwang Kim),박찬영(Chanyeong Park),이지윤(Jiyoon Lee),백준기(Joonki Paik) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
This paper proposes an optical/thermal image fusion network based on YOLOX to improve pedestrian object detection performance in night situations. The proposed network was trained and validated using the LLVIP dataset, an open benchmark for pedestrian detection under dark lighting conditions. There are two methods proposed in this paper; 1) the optical/thermal image is passed through each backbone network, and the Fusion Backbone Network (SFRM) using the Semantic Feature Receiving Module proposed in this paper is delivered to the last stage of the PAN to highlight the objects Feature Network. When performing a performance evaluation on the pedestrian object detection network in night situations, the proposed network is mAP 0.97, which is superior to other optical/thermal image fusion networks. Accurate and fast pedestrian object detection using the proposed method can reduce casualties in night autonomous driving situations because the outline of the object is not clear due to lack of lighting or light reflection.