Automatic Cameraman project (2014)

From seed
Jump to navigation Jump to search

Resources regarding HDMI recording from GoPro[edit]

Relevant Papers/Books[edit]


  1. Using high FPS camera, measure the delay between the physical movement and the display. (replicate in
  2. Ask Matt if this delay is reasonable and whether there is a simple way to reduce it.
  3. Program FPGA with echo IP (from the RIFFA public repository)
  4. Measure the latency of the echo as a function of the block size.
  5. If Delay is big, then the problem is with the PCIE.
  6. If Delay is small, we should go with the HDMI capture card.

  1. capture, record and replay from camera at (ideally at 60 FPS, minimum 30FPS)
  2. Display video in real time and with a delay of less that 100ms
  3. Add annotations to captured video, display and record.
  4. Perform simple image analysis to control the annotations.

  1. Engagement logic (implement in CPU, display through FPGA)
  2. Color based segmentation.
    • A generalized version of the skin color detector.

Student Diaries[edit]

Existing FPGA IPs from Matt[edit]


  • Skin Detector
  • Change detector
  • pop the bubble


  • Face Detector
  • Tracker


  1. Start a Diary for each person.
  2. Make the tracker work at the entrance hallway
  3. Learn the existing software:
    1. Matts Projects:
      • Are there any code repositories for the IPs and the computer software?
    2. RIFFA: new
    3. Code written by Sunsern: SunsernsDiary#Repositories
      1. Automatic cameraman (without FPGA)
      2. uRight

Steps of user interaction[edit]

  1. Engagement - finding user and making sure user agrees to be recorded.
  2. Activity - simple game, drawing program
  3. Re-Engagement in case of loss
  4. Termination

Write Up[edit]

May 2015 Writeup

Instructions for Annotating videos[edit]

  1. ssh into the hallway computer as mattj
  2. cd /home/mattj/vatic/vatic/
  3. turkic list
    1. this will give a list of videos in the system
  4. turkic find --id 'INSERT_ID_HERE'
    1. where the ID is one of the ids returned by turkic list
    2. i.e. turkic find --id 'video_03-11-2015 01:47:16'
    3. this will give a list of urls (each video is split into several parts). Go to one of the urls listed and start annotating!