Low-resolution Human Identification in thermal imagery

Object identification from an image is a tedious task in image processing. In past years many methodologies have been proposed to solve these kinds of problems. The detection of humans is paramount for applications like Advanced Driving Assistance Systems (ADAS) or Intruder Surveillance Systems or survivor detection in search and rescue efforts. Presently available image processing methods are taking a large amount of time to process the images and present an approach to identify a human object from an image which has less brightness or low resolution. Object detection system based on unsupervised learning method “K-means clustering” followed by a deep learning approach on thermal images. The system follows a two-step approach of generating anchor boxes using the clustering method and then using those anchor boxes to predict proper boundary boxes by using the deep learning method “tiny Yolo_v3”. This proposed system can suitable for real-time object detection in small-applications (less-than 50MB), embedded applications and it can achieve a much higher rate of processing when compared to traditional approaches.