A quiz on the car-counting experiment

Adrian F. Clark

1. What is the purpose of the red region drawn on the screen?

   • the bounding boxes of cars much pass through this to be counted

   • to show where the road is
   • to show where the road is
   • to distinguish lorries from cars

   The red mark on the bottom of a bounding box must pass through the red region for it to be counted.

2. How can you make the program display the individual frames more slowly?

   • use the -v qualifier

   • use the --delay qualifier

   • type Ctrl-S in the terminal window
   • you'd have to insert a call to sleep in the program

   The program allows you to specify the delay after posting a frame on the display via its -d qualifier.

3. Why is the red region not just a line?

   • it would be too narrow
   • you wouldn't be able to see it on the screen

   • a car's bounding box must pass through it to be counted

   • it would be the wrong colour

   The bounding box of a car must lie within the red region for it to be counted. Travelling very fast would mean that your car might not be counted. Note that UK speed traps don't work this way, they use radar!

4. How are cars detected?

   • they are coloured and the road is grey
   • they are large
   • they are on the road

   • they move from one frame to the next

   Cars and other vehicles are detected by their motion between frames, a very common way of identifying motion in videos. If you watch carefully, you may occasionally see that the front and backs of lorries are detected separately and this is because the sides of a lorry look much the same in successive frames.

5. How many cars are counted?

   • 71

   • 424
   • 4000
   • 100

   You mean you didn't write this down...?

6. What is the main purpose of this experiment?

   • to detect moving objects
   • to see how OpenCV processes videos
   • to count cars

   • to see how a difficult problem (identifying and counting cars) can be simplified to one of detecting crossing regions

   Although all of the answers are true, the main reason is for you to see how the overall problem, which is very difficult to solve, can be reduced to one that is much easier to solve using computer vision.

7. What would you have to do to be able to process video slips of other roads?

   • adjust the tracking threshold
   • adjust the vehicle detection code
   • adjust the global binarization threshold

   • change the information in the configuration file

   To be applied to other video sequences, all that is necessary in principle is to change the locations of the red regions in the configuration file. I am pretty doubtful that it would work on ALL video sequences though...

8. What would you need to do to count true positives, false positives etc?

   • it isn't possible to evaluate this program
   • you would need to know the correct count of cars

   • you would need to know the position of each car in each frame

   • you would need to run the video more slowly

   Alas, the only robust way to count the number of TP, FP etc is to know the location of each vehicle bounding box in each frame. That isn't impossible to do, though it could be quite time-consuming to get right.

9. What is detected in the sky towards the end of the video?

   • an aeroplane
   • a UFO
   • nothing

   • a flock of birds

   A flock of birds flies over the motorway towards the end of the video.

10. What does the configuration file contain?

    • the definitions of the red regions

    • the locations of the cars
    • the geometry of the camera and road
    • the position of the road

    The configuration file defines the red regions that appear on the screen.