Pedestrian detection based on faster R-CNN in nighttime by fusing deep convolutional features of successive images
作者：
Highlights：
• CNN training with augmented data was found effective in improving detection accuracy.
• Fusion of convolutional features in successive images enhanced detection accuracy.
• Effectiveness of our method of fusing successive-frame features is theoretically proved.
• Our analysis of complex faster R-CNN architecture helps other researchers for understanding.
• We open the trained CNN model, algorithm, and generated images to other researchers.
摘要
•CNN training with augmented data was found effective in improving detection accuracy.•Fusion of convolutional features in successive images enhanced detection accuracy.•Effectiveness of our method of fusing successive-frame features is theoretically proved.•Our analysis of complex faster R-CNN architecture helps other researchers for understanding.•We open the trained CNN model, algorithm, and generated images to other researchers.
论文关键词：Pedestrian detection,Faster R-CNN,Nighttime image,Fusion of deep convolutional features
论文评审过程：Received 12 March 2018, Revised 16 May 2018, Accepted 7 July 2018, Available online 17 July 2018, Version of Record 23 July 2018.
论文官网地址：https://doi.org/10.1016/j.eswa.2018.07.020
