Examining Transit Video Analytics

Vidient held an interesting webinar on using video analytics for transit applications (trains, subways, etc.). Read the accompanying whitepaper or access the 30 minutes webinar [link no longer available] (password - "VidientRail"). [Webinar is no longer available.]

Key takeways from the webinar:

• They are focusing on three applications: (1) track/tunnel intrusions, (2) rail yard protection and (3) abandoned object, though 1 and 2 seem to be more widely used.
• They report that they have developed/optimized an algorithm that can recognize/eliminate alerts for trains moving through (not alerting on trains is essential because of the nuisance alarm issue)
• They tune their system to handle low light inside tunnels. Following up with Vidient, they told me that they have tested their system to work down to .5 lux. As for tuning, they record sample video from the scene and when necessary adjust parameters in their video lab. I know some vendors claims zero config but I am very skeptical about zero config performance in difficult conditions like tunnels.
• Vidient's client - a US state transit agency stated that with false alarms, the 'system becomes more of a distraction than anything else" and that "false alarms are important to determine the success of the system." This is a point I emphasize repeatedly.
• In a follow up with Vidient, they recommend customers set reasonable time limits for delay in triggering abandoned objects like 10 minutes. I think this is prudent and necessary as its essentially impossible to trigger such alerts after 30 seconds without generating a flood of nuisance alarms. This may detract from its 'sex appeal' and theoretical potential but is a sound operational way to use it.