search

pageauc / speed-camera

A Unix, Windows, Raspberry Pi Object Speed Camera using python, opencv, video streaming, motion tracking. Includes a Standalone Web Server Interface, Image Search using opencv template match and a whiptail Admin Menu Interface Includes picam and webcam Plugins for motion track security camera configuration including rclone sync script. watch-app allows remotely controller camera configuration from a remote storage service name. Uses sqlite3 and gnuplot for reporting. Recently added openalpr license plate reader support.
Apache License 2.0
960 stars 169 forks source link

Frames dropped due to contour out of bounds? #154

Open carterw opened 9 months ago

carterw commented 9 months ago

I'm finding that the motion detection algorithm has trouble detecting some white-colored vehicles. A few motion events are sometimes seen but not enough to reach the MO_TRACK_EVENT_COUNT number, and then the track is terminated as having timed out. I've added a little instrumentation here and there to try to investigate.

One thing I am seeing is that some camera images are being rejected in the get_biggest_contour() method. There is code which checks to see "if object contour is completely within crop area". if x > x_buf and x + w < MO_CROP_X_RIGHT - MO_CROP_X_LEFT - x_buf:

Surprisingly, some contours are not and the image is therefore dropped. I realize the 'x_buf' number adds a fudge factor and it can be adjusted, but I don't understand how a contour "x" dimension can extend beyond the boundaries of the actual image. Can anyone explain?

More generally, why is this check necessary? What would be the effect if it were not performed? I am concerned that some legitimate motions may not be detected because the images are rejected.

SamDecrock commented 5 months ago

This function checks if the bounding box is within your self defined cropping area (configured with the MO_CROP_ variables). So increase your cropping area.

An explanation could be that the rectangle detected encompasses a piece of the background (because both the vehicle is white an so is your background?), thus reaching outside the cropping area and thus being dropped/ignored.

You can try lowering the MO_TRACK_EVENT_COUNT variable too.

carterw commented 5 months ago

The cropping area defines the image that gets evaluated to generate the bounding box. So how can the box be outside of that area?

I think part of the problem is that the motion detection algorithm always uses the X coordinate. Yes this will be the leading edge for vehicles going R2L, but in the L2R direction it is not the leading edge.