better business butler p19591
play

Better Business Butler P19591 Kollin Brakefield, Jed Katz, Marissa - PowerPoint PPT Presentation

Phase IV Gate Review Better Business Butler P19591 Kollin Brakefield, Jed Katz, Marissa McCarthy, Rory McHenry, Tom Papish, Joel Yuhas Agenda Use Cases List of Software System Block Diagram Benchmarking Functional


  1. Phase IV Gate Review Better Business Butler P19591 Kollin Brakefield, Jed Katz, Marissa McCarthy, Rory McHenry, Tom Papish, Joel Yuhas

  2. Agenda Use Cases List of Software ● ● System Block Diagram Benchmarking ● ● Functional Decomposition Test Plans ● ● Team Breakdown by Subsystem Human Subject Testing Update ● ● Web & Database Structure MSD II Preview ● ● Back End Structure Henry’s Demo Questions ● ●

  3. Use Cases

  4. Podium with Camera Wall Mounted Interface FOV: 62.2° H x 48.8° V

  5. “Hello Mr. Destler. Please follow me to your table…”

  6. Kitchen Interface It appears as if Bill Destler and Dave Munson are here

  7. System Block Diagram

  8. Functional Decomposition

  9. Team Breakdown By Subsystem

  10. Web Front End Structure

  11. Database Structure Member data will be stored in a datatable to go from the face detection backend to the web front end.

  12. Back End Structure

  13. List of Software SOFTWARE Functionality Software Webserver Node .Js Flask Facial Recognition FaceNet Microsoft API Data Storage Python SQL Pandas Detection/Image Resize OpenCV Connection TCP Interface JavaScript HARDWARE SOFTWARE MODULES Functionality Hardware Title Description Raspberry Pi Sending detections form images captured at the Camera Preprocessing RP Detection Sender Raspberry Pi camera 3b Camera Embedding Relay Picture Taking Receive a detection and send out Face ID Server System Control and Facial Server PC Interface Client Side Fetches and Displays recent profiles Recognition Manage data transmission and system logic Displaying Information Interface Ipad Phone PC Main Server

  14. Benchmarking - Camera Name Resolution [px X px] FOV [Degrees] FPS Price [USD] Dericam 1080p 1920x1080 75 30 $24.99 Logitech HP Pro C920 1920x1080 78 30 $44.99 Dericam 720p 1080x720 55 30 $19.99 EIVOTOR 720p Webcam 1280x720 ?? 30 $21.99 Logitech C270 1280x720 ?? ?? $19.99 Foscam Home Security Cam 1280x720 75 30 $39.99 ELP USB Camera 1920x1080 ~80 30 $45.99 ELP 2.1mm USB Camera 2592x1944 (maximum) ~60 8 (30 FPS Max) $43.00 ELP USB Camera for Machine Vision 2592x1944 (maximum) ~60 15 (30 FPS Max) $48.99

  15. Benchmarking - Face Recognition Max Faces in Programing Internet Connection Name Response Time (1 Face) Image Size Max Calls Photo Languages Required 2s - 6s (depends on internet and Microsoft API 64 Any 4MB (max) 20 per min Yes attributes) Facenet 1s - 2s Variable Python 200KB Infinite No Facenet will be the primary facial recognition software Long term benefits to independent and offline FR ● Low opportunity cost to switch from Facenet to Microsoft API in the event ● Facenet doesn't workout

  16. Benchmarking - Server Hardware Core Memory Graphics Card/ PSU Ram Card Bus List Product GPU Name Clock Speed Count Bandwidth Clock Speed Size Bus Width Req' Req' Port Type Price Link NVIDIA Tesla K10 745 MHz 3072 320 GB/s 5000 MHz 8 GB 512 bits 300W 16GB PCI-E x16 3.0 $178.44 Link NVIDIA Tesla K20X 732 MHz 2688 250 GB/s 5200 MHz 6 GB 384 bits 300W 16GB PCI-E x16 2.0 $99.99 Link ASUS ROG Strix RX 560 1285 MHz 1024 112 GB/s 7000 MHz 4 GB 128 bits 300W 8 GB PCI-E x8 3.0 $139.99 Link Intel NCSM2450.DK1 Movidius 600 MHz 12 400 GB/s 1066 MHz 2 MB 32 bits 7.5W 1GB USB $79.00 Link Pre-built Desktop Name CPU Speed Cores RAM Storage PSU Price Link Best-Fit GPU HP Elite 8200 High Performance SFF Business Desktop i3-2100 3.1 GHz 2 16 GB 2 TB HDD 240W $185.00 Link nVidia Tesla K10 HP Compaq Pro 6300 SFF Desktop i3-3220 3.3 GHz 2 16 GB 2 TB HDD 240W $210.00 Link nVidia Tesla K10 HP Elite 6300 Pro High Performance SFF Business Desktop i3-2100 3.1 GHz 2 8 GB 1 TB HDD 240W $146.03 Link ASUS ROG Strix RX 560 HP Elite 8300 SFF Desktop i5-3470 3.2 GHz 4 8 GB 500 GB HDD 240W $166.50 Link ASUS ROG Strix RX 560 500W Power Supply Upgrade List Price Link Comments Apevia ATX-RP500W Raptor $23.99 Link For the nVidia & Asus processors

  17. Test Plans

  18. S1 & S3 Test Plan Test Goal : Ensure the time from when a face is detected to when embedding ● and compression of image is complete is less than 5 seconds Test system's ability to ignore noise by testing with “null” subjects ● with a false-positive rate of 5% at worst Testing Parameters Sample Size = 134 6 Human Subjects - Each subject will each complete runs Neutral Facial Expression (x3) Happy Facial Expression (x3) Angry Facial Expression (x3) Hat (x3) Coat (x3) Glasses (x3) Hat, Glasses & Coat (x3) 8 Null Subjects with Plain White Paper (x2) Plain Black Paper (x2) White Paper with Multicolor Drawings of faces (x2) White Paper with Multicolor Scribbles to Represent Noise (x2)

  19. Human Subject Research Office Heather Foti confirmed on November 26th that our project “does not fit the ● federal definition of research with human subjects” Team will move forward with caution when testing with humans ● All subjects will still sign a photo release form prior to submitting to testing ● Takeaway: There is no longer an ethical risk associated with testing with human subjects

  20. MSD II - High Level Activities Activity Importance To collect key measurements and answer questions for our Henry’s Walk -through system design Purchase Parts Begin system build and integration To verify our system is designed to meet engineering Execute Tests requirements To test our system in the target environment and to collect Henry’s Proof -of-Concept Run data relevant to the handoff to the next MSD Team

  21. Henry’s Demo Questions Can we assume that all subjects will enter the floor through the main elevator? ● Where is an appropriate place to put the server? ● Can we mount a television near the POS System? ● When can we access Henry’s to record information regarding set -up? ●

  22. Questions?

  23. S2 & S4 Test Plan Test Goal: Test storage capabilities of system and ● ability to choose the correct ID amongst several potential IDs Preload system database with ID photos ● of test subjects and several other faces that are not participating in the study Test is a success if the ID of the test ● subject is returned by the system

Recommend


More recommend