Aimetis's Video Analytic Optimization Examined
Published Mar 29, 2010 00:00 AM
PUBLIC - This article does not require an IPVM subscription. Feel free to share.
Aimetis offers extensive settings / 'knobs' to optimize the performance of their video analytics. While they provide professionals strong tools to reduce false alarms, they require training and hands-on experience to master.
Aimetis provides an excellent public technical webinar that examines these points in-depth. It's the best publicly available resource on video analytic optimization we have found. It also provides a good sense of the detailed technical decisions and trade-offs that need to be made. Key optimizations examined:
- Minimum distance traveled
- Removing shadows
- Foreground sensitivty and contrast
- Remove ghost pixels (miss slow moving objects)
- merge objects
- contour-based segmentation vs people only segmentation
- track slow moving objects
- wait after large background change before starting to track objects
- reduce false tracks from snow
- smallest object size (pixels)
- background learning duration in seconds
- check for sudden background movement
[Note: Aimetis allows users to choose from 9 video analytic algorithms. This is for their VE150 algorithm, optimized for outdoor use.]
I'd encourage you to watch the webinar as they do a good job of honestly explaining the tradeoffs.
As they acknowledge at the end of the webinar, their default settings are optimized to almost never miss alarms (but are tolerant of triggering false alarms). These optimizations are for those who want to lower false alarms while accepting some missed valid alarms and/or delays on alarming.
An interesting debate can be had on whether one needs these optimization settings or simply should use intelligent defaults. While the intelligent defaults approach (like VideoIQ) can reduce setup time, Aimetis's advanced optimizations would help rectify many of the most common false alarms we found in our VideoIQ test.