/speed_detection

Primary LanguageJupyter Notebook

Speed Detection Repository

Peronal project inspired by the Comma AI speed challenge: github.com/commaai/speedchallenge

Your goal is to predict the speed of a car from a video.

  • data/train.mp4 is a video of driving containing 20400 frames. Video is shot at 20 fps.
  • data/train.txt contains the speed of the car at each frame, one speed on each line.
  • data/test.mp4 is a different driving video containing 10798 frames. Video is shot at 20 fps.

Installation instructions:

  • If necessary, pip3 install --upgrade pip
  • clone this repo
  • download data from the linked repo
  • make create_environment
  • conda activate speed_detection
  • make requirements (adjust Makefile if not Cuda 10.2)

Please run nbs/data_processing to set up data, which also runs a segmentation network over all data to be used for image masking. Finalized notebooks: fixed_point_regression, SuperPoint_fixed_point_regression.

Our approach ultimately uses SuperPoint for feature point tracking, and normalizes the found vectors that appear only on the road in front of the car. A regression model is fit to the median of these tracks for each image pair. No classical image processing algorithms are needed such as optical flow or corner detectors.