PhD Dissertation Proposal by Amit | Vision-based Analytics for Improved Real-World AI-driven IoT Applications

About The Talk

The market for Internet of Things (IoT) devices is witnessing rapid growth in recent times, primarily because of low-cost sensors and emergence of powerful embedded device platforms. Vision based sensing via such IoT devices has enabled a wide variety of compelling new applications across domains such as home automation, smart healthcare, smart appliances and smart city analytics. These applications are being driven by the introduction of Deep Neural Network (DNN) based models, which enable capabilities such as object detection, face recognition and human activity recognition. Nevertheless, the use of such DNN-based visual processing, over data collected by vision sensors embedded in pervasive devices, still has several limitations that inhibit their effective use for in-the-wild deployments, across locations such as residential homes, university campuses and retail & wholesale markets. This thesis explores how several such limitations for real-world, AI driven IoT-based applications can be overcome by advances in multiple dimensions, including (a) using additional sensing modalities to better isolate objects that need to be visually sensed (b) exploiting multi-camera collaboration to reduce sensing energy overheads and (c) combining multiple, noisy partial views of object of interest to derive holistic attributes of the object.

Speaker Biography

Amit is a PhD candidate in the School of Information Systems, Singapore Management University, working in the area of mobile and ubiquitous computing. He received his M.Tech in mobile & ubiquitous computing degree from IIIT Delhi in 2014. His current research focuses on designing and developing vision-sensing systems for improving Internet of Things (IoT) applications.