Wednesday, September 19, 2018 - 3:00pm to 4:00pm
Event Calendar Category
LIDS & Stats Tea
University of Cambridge
Building and Room Number
The cost of modelling existing industrial facilities is currently considered to counteract the benefits of the models in maintaining and retrofitting a facility. 90% of the modelling cost is typically spent on labour for converting point cloud data to the 3D model, hence reducing the cost is only possible by automating this step. Previous research has successfully validated methods for modelling specific object types such as cylinders. Yet modelling is still prohibitively expensive. During this talk, the most important industrial object types will be identified by ranking them according to their frequency of appearance and the man-hours required for modelling in state-of-the-art modelling software. This work is the first to rank objects according to their priority for automated modelling. These objects are in descending order of frequency of appearance: electrical conduit, straight pipes, circular hollow sections, elbows, channels, solid bars, I-beams, angles, flanges and valves. We present CLOI (Channels, L-shapes, circular sectiOns, I-shapes): a richly-annotated large-scale repository of shapes represented by labelled point clusters. CLOI organizes the labelled clusters in the 10 most frequent categories encountered in industrial plants for all the facilities. It has more than 10,000 labelled clusters that can be directly used in deep learning applications. Automated detection and semantic classification methods for the recognition of the abovementioned objects will be analyzed.
Ms. Agapaki is a Charles M. Vest Scholar at CSAIL and 3rd year Ph.D. Candidate at the University of Cambridge. She is advised by Dr. Ioannis Brilakis (University of Cambridge) and Dr. Justin Solomon (MIT CSAIL). She holds a Master of Science in Civil Engineering from the University of California, Los Angles (UCLA) and a BS in Civil Engineering from the University of Patras in Greece. Her Ph.D. research focuses on the automated generation of as-is geometric Building Information Models (BIMs) of industrial facilities. The methods that she investigates combine deep learning methods for point cloud data and engineering knowledge for the layout of the piping and structural system of the most important industrial objects. Eva won the first prize in the Innovation Competition of Lean and Computing in Construction Conference last summer and the “Rising Star” Poster Award last year in Oxford during the 4th Oxbridge Women in Computer Science Conference.