5.0.3 - New Edge Vector Motion Algorithm

fenderman

Staff member
Joined
Mar 9, 2014
Messages
36,897
Reaction score
21,250
5.0.3 - September 15, 2019
With a continued focus on AI for version 5, it was time to upgrade one of the most critical of
algorithms in the software—the motion detector. You will find a third choice has been
added to the algorithm selection box found on the Motion configuration page from the
Trigger tab in camera settings—Edge vector. This will eventually replace simple as the default.
The algorithm discerns between leading edge and trailing edge motion and uses this
information to compute a motion vector (magnitude and angle). The goal for this new
algorithm is to reduce false positives (a consistent vector is required over the make time in
order to trigger) and to feed more advanced AI with more relevant frames. Please provide
feedback on effectiveness (in the form of BVR video clips for analysis)


More detail from the help file.

You may select to use either a Simple, Gaussian, or Edge Vector Algorithm. The Gaussian
algorithm uses slightly more complex heuristics for tracking pixel changes over time,
possibly helping to reduce false positives, but at a slight increase in CPU demand.
The simple algorithm emulates version 4, but the newer and somewhat more sophisticated
edge vector algorithm soon will become the default for version 5. This new algorithm
distinguishes between the leading and trailing edges of motion and you will see this if you
enable a highlighting option either in camera settings or in the viewer for testing. The
leading edge is painted a bright shade of blue, while the trailing edge appears in orange.
The remainder of the movement is a darker blue as was used in the simple algorithm.
The algorithm uses the leading and trailing edges to compute a vector, which consists of a
magnitude and angle for the motion of the object. A trigger will only occur if this vector is
consistent for the duration of the make time.
 

Forid200

Young grasshopper
Joined
Jul 11, 2019
Messages
31
Reaction score
7
Location
London
Really looking forward to testing this! Been suffering with false positives a lot.
 

Wippie

n3wb
Joined
May 27, 2017
Messages
4
Reaction score
2
Yes, we need more AI in order to prevent false positives, especially with wind/trees/sun combination. I'm testing Sighthound Video which has far less functionality but works much better in my case (no false positives that is).
 

Forid200

Young grasshopper
Joined
Jul 11, 2019
Messages
31
Reaction score
7
Location
London
More important than false positives is false negatives.
You're absolutely correct! Hence why I've been putting up with a dozen alerts from headlights at night.

I'm also patiently waiting for a solution, I've tried some AI solutions with BlueIris but I could never get them to either work consistently, or they give false negatives which can be rather worrying.
 

pmcross

Pulling my weight
Joined
Jan 16, 2017
Messages
371
Reaction score
185
Location
Pennsylvania
I updated BI and changed the motion algorithm from simple to Edge Vector on 2 of my cameras. I will update everyone in a day or so to see if false positives have declined.
 

J Sigmo

Known around here
Joined
Feb 5, 2018
Messages
997
Reaction score
1,333
If this does reduce the false positives from tree shadows in the wind, and car headlights sweeping my yard and porch, that certainly will be a big plus.
 

pov2

Getting the hang of it
Joined
Sep 7, 2018
Messages
229
Reaction score
46
Location
Canada
On 5.0.3.2 I changed from Gaussian to this new Edge Vector and stopped receiving any motion alerts even the legitimate ones where people cross zones, for which I always received alerts. I went to change back to Gaussian and was greeted with a message that new version 5.0.4.0 is available with a new Edge Vector algorithm. Updated. Will see.
 

TL1096r

IPCT Contributor
Joined
Jan 28, 2017
Messages
1,223
Reaction score
465
On 5.0.3.2 I changed from Gaussian to this new Edge Vector and stopped receiving any motion alerts even the legitimate ones where people cross zones, for which I always received alerts. I went to change back to Gaussian and was greeted with a message that new version 5.0.4.0 is available with a new Edge Vector algorithm. Updated. Will see.
I was thinking of making a clone of one camera and use new edge vector with one and Gaussian with the other. Going through the motion detection would determine what is missed and picked up between the 2 settings.
 

pov2

Getting the hang of it
Joined
Sep 7, 2018
Messages
229
Reaction score
46
Location
Canada
Not better with the 5.0.4.0. Goes back to Gaussian.
 

IAmATeaf

Known around here
Joined
Jan 13, 2019
Messages
3,287
Reaction score
3,252
Location
United Kingdom
Same for me, changed 2 cams and no recordings since making the change, changed then back now.
 

Wippie

n3wb
Joined
May 27, 2017
Messages
4
Reaction score
2
I changed it too with great success it seems, I no longer have false positives, and it seems no false negatives either (running Blueiris and Sighthound in parallel so I can compare). Updated to 5.0.4.
I changed several settings when I set New Edge Vector, for now: Make at 0.6 seconds, Cancel shadows, Object travels 200 pixels and crosses zone A, Minimal object size setting, Minimal contrast setting.
With or without sun, both people and vehicles are triggered as they should.
I'm waiting for more wind and clouds though for further assessment...
 

pov2

Getting the hang of it
Joined
Sep 7, 2018
Messages
229
Reaction score
46
Location
Canada
It looks like one zone (A) is needed for this Edge Vector algorithm. It should cover the whole area where motion detection is desired. Because of my previous struggles with shadows etc. I created multiple zones on one cam and set complicated rules when to trigger an alert. This cam stopped sending alerts when obvious motion occurred. I didn't care much about another cam and it had just one large zone A. This cam triggered alerts with the Edge Vector, and when I looked at the videos it looked like a large area was needed to "create a vector". So the first cam didn't have a chance with small multiple zones. The bad news - I got a lot of false positives from car lights during the night from the second cam with Edge Vector.

Does anybody feel the same way about the requirement of a single large large zone for Edge Vector?
 

Wippie

n3wb
Joined
May 27, 2017
Messages
4
Reaction score
2
It looks like one zone (A) is needed for this Edge Vector algorithm. It should cover the whole area where motion detection is desired. Because of my previous struggles with shadows etc. I created multiple zones on one cam and set complicated rules when to trigger an alert. This cam stopped sending alerts when obvious motion occurred. I didn't care much about another cam and it had just one large zone A. This cam triggered alerts with the Edge Vector, and when I looked at the videos it looked like a large area was needed to "create a vector". So the first cam didn't have a chance with small multiple zones. The bad news - I got a lot of false positives from car lights during the night from the second cam with Edge Vector.

Does anybody feel the same way about the requirement of a single large large zone for Edge Vector?
Absolutely!
 
Joined
Apr 20, 2018
Messages
28
Reaction score
13
In my limited testing I’ve noticed I used to have black and white turned off to help with shadows but with edge vector it seems I have to turn it back on to help. Didn’t test much but was something I noticed.
 

Walrus

Getting comfortable
Joined
Nov 19, 2018
Messages
593
Reaction score
449
Location
Ontario
I haven't had a sunny day with fast moving clouds to see the results, but it has caused BI to miss a lot of motion. It only picks up my cars coming into the driveway half the time now. I've even reduced the make time to 0.5s (was 1s), min object size 200 (was 250), object moves 200 (was 300), and my contrast is set to 20. Object size exceeds is 67%. Yet it misses an entire car.
 

Mikk36

Getting the hang of it
Joined
Aug 21, 2018
Messages
105
Reaction score
42
Location
Estonia
Better stock up on some sample clips to keep testing the settings (with "test run video through motion detector").
 

pinko

Known around here
Joined
May 26, 2017
Messages
385
Reaction score
623
Location
Australia
Testing on two front gate cams Simple vs EV. Not much activity since update earlier today.
Might need one of those dreaded windy + cloudy days to make some comparisons & adjustments.
Ken is always on top of his game. It's hard not to be amazed/ impressed by his efforts.
 

jaydeel

BIT Beta Team
Joined
Nov 9, 2016
Messages
1,126
Reaction score
1,237
Location
SF Bay Area
Might need one of those dreaded windy + cloudy days
Just for fun I've started calling these conditions "gleamy" after coming across this interesting website ...
Long Forgotten Weather Words

GLEAMY... "... the weather is gleamy [when] it's intermittently sunny, or as one nineteenth century glossary put it, "fitful and uncertain."
CARRIES... "Cairies are swiftly moving clouds. An old Scots dialect word, it derives from cairy (a Scots pronunciation of "carry"), a local name for a burden or a load to be conveyed.
 
Top