University of Twente Student Theses


Human Detection in aSequence of Thermal Images using Deep Learning

Wang, Xinran (2019) Human Detection in aSequence of Thermal Images using Deep Learning.

[img] PDF
Abstract:Human detection technology plays an irreplaceable role in many areas, such as search and rescue(SAR), autonomous driving, and surveillance,in recent years.Human detection isastill challengingtask because,for the group of people, each individual has his unique appearance and.body shape. At the same time, humans can make thousands of gestures.Compared with the traditional method, the deep learning neural networkhas the advantages of shorter computing time, higher accuracy and easier operation. Therefore, deep learning method has been widely used in object detection.The current state of art in humandetection is RetinaNet. It is a robustone-stage object detector (Lin, Goyal, Girshick, He, & Piotr Dollar, 2018).This approachproposed a new functionof loss to address the imbalance between foreground and background classes.The temporal component of video provides additional and significant clues as compared to the static image. In this paper, the temporal relationship of the images is utilizedto improve the accuracy of human detection. Compared to using only an image, the accuracyof human detection is higher when a sequence of images is applied.The dataset used in this thesis is from KAIST. The dataset is disappointing because it hassuffered a lot of occlusion and wrong annotations.In addition, the image resolution is low, and theldistance between people and the camera makes the outline of people very vague.That will have anegativeimpact on the result.In this study, three temporally-consistent convolutional neural networks and a basic convolutional neural network have beenusedto do human detection. Andthe results show that all temporally-consistent CNNs perform better than the basic CNN.The continuous later CNNhas the best performance that the accuracy ofhuman detection is 21.4% higher than that of the basic model.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page