Object Recognition on Long Range Thermal Image Using State of the Art DNN

This paper presents a method for detecting objects on long range thermal images. This problem is extremely complicated due to the low resolution of these images, and due to the fact that the objects that we want to detect have areas as small as 50 pixels. In order to solve this problem we have developed, trained and tested a large number (2640) of variations of the state of the art algorithm for visible camera object detection, called YOLO Darknet - You Only Look Once Darknet. The paper succeeds in showing that the deep neural network YOLO Darknet can be adapted to work on thermal images and, furthermore, it proves that the network is capable of detecting and classifying objects that are hard to detect with the human eye.

[1]  Ali Farhadi,et al.  YOLO9000: Better, Faster, Stronger , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[2]  Bogdan Ionescu,et al.  Ensemble-based learning using few training samples for video surveillance scenarios , 2015, 2015 International Conference on Image Processing Theory, Tools and Applications (IPTA).

[3]  Irfan Riaz,et al.  HUMAN DETECTION BY USING CENTRIST FEATURES FOR THERMAL IMAGES , 2013 .

[4]  Weibin Liu,et al.  Robust motion detection using histogram of oriented gradients for illumination variations , 2010, 2010 The 2nd International Conference on Industrial Mechatronics and Automation.

[5]  Bogdan Ionescu,et al.  Little-Big Deep Neural Networks for Embedded Video Surveillance , 2018, 2018 International Conference on Communications (COMM).

[6]  Jenny Benois-Pineau,et al.  Segmentation-based multi-class semantic object detection , 2012, Multimedia Tools and Applications.

[7]  Little-Big Deep Neural Networks for Embedded Video Surveillance , 2018, 2018 International Conference on Communications (COMM).

[8]  Ah Chung Tsoi,et al.  Face recognition: a convolutional neural-network approach , 1997, IEEE Trans. Neural Networks.