IoTracker: Home-made Tracking System using Core-sets and the Internet of (Tracking) Things Soliman Nasser Ibrahim Jubran Artem Barger Dan Feldman Robotics and Big Data Lab, Department of Computer Science , University of Haifa IoTracker A low-cost motion tracking system for indoor localization, based on web-cameras connected to “Internet of Things” (IoT) boards. • 40 FPS real time tracking. • Full 6DOF (𝑋, 𝑌, 𝑍, 𝜑,𝜃,𝛹) • Low-cost hardware. • Easy to deploy. • Mobility. System Overview The hardware of our system consists of the following commercial products: • Web camera: A ~10$ camera, used for tracking, both in IR and RGB modes. • Client’s board: Odroid U3, or Intel’s Galileo Gen2 (30$-50$). Each camera is connected to its own board using a USB cable. Each board is equipped with a wifi adapter for communication to the server. • Server’s board (same as Camera’s board). This is the single board that collects the tracking information from all the other cameras through the UDP based communication protocol that we implemented. Problem Statement Perspective-n-Point: The aim of the Perspective-n-point problem is to determine the position and orientation of a camera given its intrinsic parameters and a set of n correspondences between 3D points 3D lines. Our Approuch Coreset for PnP: Definition: An 𝜺-coreset for PnP is a set 𝑆 ⊆ 1,2, … , n of 𝑆 indexes and 𝑆 weights 𝑤1 , 𝑤2 , … , 𝑤𝑛 such that: 𝑂𝑃𝑇 𝑃, 𝐿 is approximated by min 𝑖∈𝑆 𝑤𝑖 ∗ dist 2 (𝑇 𝑝𝑖 , 𝑙𝑖 ) up to (1-𝜀) multiplicative factor. 𝑇∈𝑇 Formal Definition of the PnP: Let P = {𝑝1 , 𝑝2 , … , 𝑝𝑛 } be a set of n points, and 𝐿 = {𝑙1 , 𝑙2 , … , 𝑙𝑛 } be a set of n lines, both in 𝑅3 . Compute a translation and rotation 𝑇 over every rotation and translation 𝑇 in 𝑅3 , that minimizes: OPT P, L : = 𝑛𝑖=1 dist 2 (𝑇 𝑝𝑖 , 𝑙𝑖 ), where dist 2 𝑝, 𝑙 is the squared distance from a point 𝑝 to the line 𝑙. Figure 7: A set of 3D point (P), and a subset of points {𝒄𝟏 , 𝒄𝟐 , 𝒄𝟑 , 𝒄𝟒 } ∈ 𝑷 which are the coreset of P Figure 1: IoTracker clients installed near the ceiling of Jacob’s building, University of Haifa. Motivation Motion Tracking System. Motion tracking systems can be used for indoor localization, navigation, and robot controlling. Applications: • Guardian Angel: navigation people inside buildings. (malls, hospitals, etc.) • Autonomous drones and vehicles. • Smart Homes. Existing Systems. Existing motion capture systems for such application use dedicated hardware and workstations that cost thousands of dollars, and thus exist mainly is research labs. Our system will give this ability to non-experts that only have a limited budget. Figure 5: 5 pairs of points and lines correspondences in 𝑹𝟑 , from which we calculate the translation and rotation matrices. Figure 3: System overview, Left: server side, Right: client side Camera Hardware Challenge We needed a fast and simple detection and tracking algorithm, therefore we disassembled and modified the camera lens in order to capture only IR wavelengths. Streaming version of the PnP: Definition: Update 𝑂𝑃𝑇(𝑃, 𝐿) after insertion / deletion of a pair (𝑝𝑖 , 𝑙𝑖 ) of point 𝑝𝑖 in 𝑃 and it’s corresponding line 𝑙𝑖 in 𝐿 using 𝑂 logn time and memory per pair. Our research challenge is to prove the following theorem: For every 𝜀 > 0 there is an 𝜺-coreset (𝑆, 𝑊) for (𝑃, 𝐿) of size logn 𝑆 =𝑂 . This coreset can be updated in 𝑆 time per ε insertion / deletion of a pair 𝑝𝑖 , 𝑙𝑖 of point 𝑝𝑖 and line 𝑙𝑖 References [1] Lepetit, Vincent, Francesc Moreno-Noguer, and Pascal Fua. "Epnp: An accurate o (n) solution to the pnp problem." International journal of computer vision 81.2 (2009)” [2] S.Nasser, A.Barry, M.Doniec, G.Peled, G.Rosman, D.Rus, M.Volkov, D.Feldman. Fleye on the Car: Big Data meets the Internet of Things.” ACM 14th International Conference on Information Processing in Sensor Networks (IPSN '15)” Figure 4: Left: A picture from a pre-modified camera, Right: A picture from a modified camera, capturing only IR wavelengths Figure 6: Features selected, on the robot, for the PnP algorithm [3] D.Feldman, M.Langberg. A Unified Framework for Approximating and Clustering Data. “ACM Symposium on Theory of Computing (STOC 2011)” Figure 2: Autonomous mini quadcopter hovering in Jacob’s building. RESEARCH POSTER PRESENTATION DESIGN © 2012 www.PosterPresentations.com