Skip to main content

Capturing videos from traffic cams

Capturing videos from Seattle using m3u8 playlist file​

  1. Go to this website SDOT Travelers Home Page (seattle.gov)
  2. Choose a camera for which you want to capture the video feed
  3. Click on the video button and then using the Developer tools, find the GET URL for the first playlist file
  4. !ffmpeg -i "Enter link ending in m3u8 over here" -c copy -bsf:a aac_adtstoasc "output_file_name.mp4"
  5. Using the timeout function, one can download the video from the playlist for any desired duration of time.
  6. timeout 1m ffmpeg -i "Link" -bsf:a aac_adtstoasc -vcodec copy -c copy -crf 50 <file_name>.mp4

Capturing images from Cincinnati traffic cameras​

  1. Go to the website OHGO
  2. Choose a camera for which you want to capture the images
  3. The website does not have a live video stream, it refreshes the images every 5 seconds
  4. Click on the camera button and then using the Developer tools, find the GET URL for the image file
  5. Feed these links to get_traffic_cam_images.py file
  6. The script then downloads images at an interval of every 5 seconds

Capturing images from flood dynamically​

  1. Host the python file on Github and create a simple front end with a button and desired duration to capture the images
  2. Whenever it starts raining, open the webpage and press the button and select the required duration to start capturing the images

Background subtraction​

  1. Feed the video that you want to subtract background from in the video source tab
  2. Then you can see the source code
  3. Create a CLI version of capturing Cincinnati images
  4. It is created and a URL can be passed

Azure:

  1. Bing Web search API allows you to download images based on a query in bulk. The free tier allows up to 1000 queries and the details for the paid tier are available over here. Bing Search API Pricing | Microsoft Bing.
  2. The paid tier has various options going from $3 to $7 per 1000 transactions. Each transaction allows us you to download 150 images in a single click

3D modelling from images of cars from traffic cameras​

  1. Paper sent by Sudhir, goes over 2 perspectives, first still object and moving camera, the 3D model is computed using the triangulation and 2nd one being moving object and moving camera.
  2. It does not discuss the approach for moving objects and still camera. The task is search for implementations where from a given traffic camera, with 2 or more snips of a moving car can a 3D model be constructed?
  3. If so, can that 3D model be used for cluster analysis? Is the comparison of results from the cluster analysis of 3D images and 2D images possible? Which one would perform better?
  4. Approach discussed during the meeting was the use lazy suzan for 3D modelling, where an object say a car is rotated on a turn table like structure extremely slowly and multiple images are captured. With all the images being captured, careful reconstruction of the 3D model is possible using the images captured.