2016
Gregoire Robinson
I am a rising junior at the University of Massachusetts Amherst pursuing a B.S. in Computer Science. This summer, I am working with Dr. Shah and Dong Zhang on detecting moving vehicles on roads and highways for Wide Area Surveillance (WAS). These images are obtained from a large electro-optic camera system flying at ~7000 ft. This is a challenging problem because the images are low resolution, the vehicles have a low saliency from the image background, image capture frequency is low, and there are potentially thousands of vehicles in every image. Current frame difference and saliency detection methods are designed to use appearance and motion information separately and are therefore not very robust. Our approach aims to circumvent these limitations using deep convolutional neural networks (CNN) and long short-term memory (LSTM) networks. By feeding multiple consecutive frames into the networks, the goal is to train them to learn both appearance features and multi-frame frame difference cues. This technology has applications in security, traffic monitoring, and emergency management. Check out the weekly presentations below to learn more about the project.
Weekly Presentations
Poster Presentation
PDF
Final Report
PDF
Back to 2016 REU Participants