What is an algorithm I can use to program an image compare routine to detect changes (like a person coming into the frame of a web cam)? -

i have web cam takes picture every n seconds. gives me collection of images of same scene on time. want process collection of images created identify events entering frame, or else large happening. comparing images adjacent in time , fixed in space - same scene @ different moments of time.

i want reasonably sophisticated approach. example, naive approaches fail outdoor applications. if count number of pixels change, example, or percentage of picture has different color or grayscale value, give false positive reports every time sun goes behind cloud or wind shakes tree.

i want able positively detect truck parking in scene, example, while ignoring lighting changes sun/cloud transitions, etc.

i've done number of searches, , found few survey papers (radke et al, example) nothing gives algorithms can put program can write.

1. use color spectroanalisys, without luminance: when sun goes down while, similar result, colors not change (too much).

2. don't go big changes, quick changes. if luminance of image changes -10% during 10 min, means usual evening effect. when change -5%, 0, +5% within seconds, quick change.

3. don't forget adjust reference values.

4. split image smaller regions. then, when regions change same way, know, it's global change, eclypse or what, if 1 region's parameters changing, happens there.

5. use masks create smart regions. if you're watching street, filter out sky, trees (blown wind), etc. may set different trigger values different regions. regions should overlap.

6. a special case of region line. line (a narrow region) contains less , more homogeneous pixels flat area. mark, say, green fence, it's easy detect wheter crosses it, makes bigger change in line in flat area.

7. if can, change irl world. repaint fence strange color create color spectrum, can identified easier. paint tags floor , wall, can ocred program, can detect wheter hides it.