(Dis)occupancy

University: Institute for Advanced Architecture of Catalonia
Masters in Robotics and Advanced Construction
Class: Sensing Machines Studio

The population in Barcelona has almost doubled over the past 50 years, however the city size remains quite unchanged. How can we then accommodate for the urban infrastructure required by a modern and growing society? We decided to focus on human mobility as it is one of many factors dictating city planning decisions, so we set out to study which spaces humans are occupying, and more importantly which spaces they aren’t. This project aimed to tackle this problem by using computer vision to detect, track and record pedestrian movement in a city, and introduce urban utility where unoccupied spaces could be found.

​

• Utilized the YOLOv4 Convolutional Neural Network for real-time object detection w/ DeepSORT for identification & tracking.
• Recorded pedestrian paths & used satellite data to homographize those coordinates & generate a 2D occupancy grid.
• Generated a 3D point cloud and mesh of the space using photogrammetry and LIDAR scanning.
• Point Cloud segmentation to obtain ground and walls using Metashape & Open3D in python.
• Created an algorithm in Grasshopper to populate voids with urban furniture and greenery.