The Montreal Metro is Canada's second busiest subway system, and North America's fourth busiest in total daily passenger usage, delivering an average of 1,241,000 daily unlinked passenger trips per weekday. The Metro is operated by the Société de transport de Montréal (STM), Montreal’s metro transit authority. STM also operates one of North America's largest urban bus and rail rapid transit schemes, serving the third-largest number of passengers overall behind New York and Mexico City, and attracting the second-highest ridership per capita behind New York
In 2008, STM sought to improve public safety by automating its detection of people intruding on the tracks and into the tunnels of the Metro system. STM already had a Loronix Windows-based software management system running its video surveillance system, with more than 300 of the several thousand cameras viewing the Metro tracks and tunnels. The next logical step for STM was to migrate to an intrusion detection solution using video analytics software that also ran on Windows servers that would fit well into its existing infrastructure and information technology skills sets.
Successful deployment of a video-based intrusion detection system on such a large scale is difficult in a metro rail environment, where camera views are often filled with the intense motion and illumination variations of a train. Such conditions usually render motion-based video analytics temporarily useless.
Vidient Systems of California, however, had successfully deployed its SmartCatch video analytics software running on Windows servers for rail intrusion detection on a small scale. SmartCatch already had advanced capability to reliably recognize a walking or running human being – even with only 60-70 pixels on the target. Since Vidient had committed to solving the problem of dominant train motion, STM contracted them to help solve the issue at the Metro.
STM was well aware that the performance of video analytics for intrusion detection is sometimes a trade-off between the probability of correctly detecting an intrusion and the rate of false alarms. STM realized that too many false alarms would cause operators to lose confidence in the SmartCatch solution, and even cause some operators to turn the system off.
Many video-based intrusion detection installations can tolerate up to one false alarm per camera per day. But with almost 300 cameras running video analytics in the Montreal Metro, a much lower false alarm rate was essential. Even if these cameras were generating one false alarm per camera per day, there would be a flood of false alarms -- one every 4-5 minutes on the average – perhaps worst at busy train times because train motion is the primary trigger of false alarms. This would be an intolerable false alarm rate, considering STM needed to maintain at least a 90 percent-plus correct detection rate.
Normal operations in Montreal generate about 200 authorized intrusions by work crews every day, giving a very useful ongoing check of sensitivity. The Montreal Metro authority required that there be no more than 0.5 false alarms per camera per day – about 150 for the whole system. Thus correct detections would at least be noticeably greater than false alarms.
For almost a year, Vidient and STM cooperated in the development of improved SmartCatch software. The goal was to successfully detect a human intruder on a catwalk next to a moving train that filled most of the camera’s view, while generating a minimal number of false alarms. The final hurdle came during a week-long acceptance test in August, 2009, and again in the summer of 2011. The result demonstrated a false alarm rate of 0.15 to 0.20 false alarms per camera per day. The ratio of correct detections to false alarms was at least 5 to 1, providing the systems operators the confidence they needed as the probability rate averaged approximately 95 percent.