These days spending off time to work on vehicle detection project - my task is to recognize a
standing vehicle at checkout for a local customer in RWP.
Significant material is available for recognizing moving vehicles using optical flow, frame differences and GMM techniques etc. For
static vehicles, people suggested fix video background to perform subtraction techniques. I tried but didn't get accuracy due to
variations in camera capture conditions specially illumination.
Eventually after sepeding some time, Speeded-Up Robust
Features algorithm with significant points worked better than others to detect a stationary
vehicle based on measuring average object size, appearance duration and movement
patterns.
See this detection video: https://1drv.ms/v/s!AnZugs0jUgZhg2e7eU-lEmePK6Hl
Couple of challenges are yet to solve. Though It detects non-overlapping
vehicles almost 96% correctly. Need to solve overlapping vehicles very close to each other
appear to camera as single object, then bikes as non-target objects are decteed as false positive.
See challenges video: https://1drv.ms/v/s!AnZugs0jUgZhg2i2I8NOnD23igsL
See challenges video: https://1drv.ms/v/s!AnZugs0jUgZhg2i2I8NOnD23igsL
Thumbs up :)