Scene understanding using deep learning

Abstract

Automation based on artificial intelligence becomes necessary when agents such as robots are deployed to perform complex tasks. Detailed representation of a scene makes robots better aware of their surroundings, thereby making it possible to accomplish different tasks in a successful and safe manner. Tasks that involve planning of actions and manipulation of objects require identification and localization of different surfaces in dynamic environments. The usage of structured-light-based depth-sensing devices has gained much attention in the past decade. This is because they are low-cost and capture data in the form of dense depth maps, in addition to color images. Convolutional Neural Networks (CNNs) provide a robust way to extract useful information from the data acquired using these devices. In this chapter we discuss the basic idea behind standard feedforward CNNs, and their application to semantic segmentation and action recognition.