“We’re making a lot of headway in the transit space,” said Wright. “They have a lot of cameras which create a big problem with a lot of data and they don’t have the human resources to put eyes on all of them. But we bring those exceptions to the top so that they can intervene and respond when something highly unusual or dangerous happens.”
With the learning process of video analytics, the system observes everything that happens in the camera and like the human brain, the path of the memory of what is occurring. It builds a model of behavior and everything that it witnesses, such as the environmental changes like sunrises, shadow casting, birds flying or wind blowing trees. It also learns the people behavior patterns of pedestrian traffic, the normal trajectories they take and the rate of travel and vehicular patterns, which include the speed and trajectory of the train and any other types of vehicles in the sight.
Critical Solutions International Project Engineering Manager Michael Williams talked about how they set up live alarms for video analytics and how it can be used for rapid historical searching. While certain things can trigger alarms, archive video data can be searched for categorized data.
Wright stated that there have been people driving cars on the track when they’re not allowed to be there. “It’s a very dangerous thing so we’ll learn … it’s normal for trains at the time but we’ve never seen a vehicle like this in this area, it’s unusual or an anomaly and we’ll generate an alert on that.”
And TriMet has had experience with that type of intrusion. Wilkinson said, “We’ve had a couple of intrusions … with vehicles down our alignment off of service streets where we actually put big concrete blocks.” They’re on top of the existing ties and make a large speed bump so as people hit the barriers, it stops the cars before they get too deep into the line where it’s very difficult to get them out.”
He said of one instance, the car made it from 16th Avenue all the way up to 42nd Avenue and by that point, it was very difficult to get them out of the alignment. It required having a contractor go in there with a crane to get the car up and out.
Analytic system Set Up
The time it takes to set a system up and getting it operating depends on a variety of factors Williams said. “If we have a typical subway station that may have 20 to 30 cameras on it, you probably don’t want to set up alarms on all 30 cameras.
“If you want to that’s fine, but usually you can cover most of your area with a few cameras with the alarm.” He also said it does take some time to calibrate the system to ensure there are as few false positives as possible.
Wright said a general guideline is about two weeks for getting a base-level understanding, enough of a model created to alert. However, he reiterated, the system is continuously learning and is constantly adding to the modeling.
Wright said the term “false positives” is thrown around some and in reality, everything alerted is in fact a visual anomaly, but it’s a matter of whether those anomalies are truly valuable to the customer.
Williams mentioned some of the changes that naturally can occur which will create initial anomalies, such as a change in surroundings that creates new shadows, and those are things they research and find ways to continually improve.
“Sometimes people are afraid of the analytics,” said Williams. They think it’s maybe trying to replace the human side so some people are worried about their jobs.
“We certainly don’t want to replace anyone,” he said, “but the idea is that the person there, they become more efficient.”
“Generally the transit agency that we’re working with have a lot of cameras,” Wright said. “You’re talking hundreds or thousands of cameras.” Often the agencies are getting grants to install the cameras. If you go get millions of dollars to put out hundreds or thousands of cameras and you don’t have a plan to manage video, view the video or make use of it, then you may not get the grants to install the camera.”
He stressed, “The key is — and needs to be — technology that doesn’t create more problems than it solves.”