Best Practices: The Importance of Detecting Intruders and Obstacles on High-Speed Transit Tracks

Nov. 19, 2020
Integrating Deep Learning Neural Network technology into detection systems can boost the accuracy of these systems and reduce the risk of train collisions.

In today’s world of high-speed transportation, reliance upon mass transit systems continues to increase. As countries build and expand an ever-increasing number of transit rail systems to interconnect cities and suburbs, the importance of safety and security of these transportation systems also increases and the consequences of their failure can result in injuries, loss of life and major financial impacts.

There are many potential threats to transit rail systems. This article focuses on rail intrusion detection technologies, which are able to ensure:

  1. That tracks are clear of obstructions;
  2. That those attempting to cause infrastructure and collateral damage are thwarted; and
  3. That actionable, trustworthy and timely intelligence is delivered to train operators for appropriate action.

The types of objects that can obstruct trains include large and smaller vehicles, people, large animals, cargo that falls from cargo rail trains and damage to infrastructure. The presence of these obstructions can be either accidental (e.g. stalled vehicle at an intersection or a person falling from a station platform) or intentional (e.g. suicide attempt or a terrorist event).

Generally, individuals can gain access to transit tracks at stations, intersections or wherever there is a lack of adequate barrier fencing. Animals, such as deer, may even be able to leap over and individuals may be able to climb over fences and obstruct tracks. Vehicles can access tracks via road/rail intersections. Boxes, crates and other items that accidentally fall from cargo trains can end up on the tracks and present a serious hazard to approaching trains. A terrorist might attempt to damage a bridge, section of rail or a tunnel to cause serious loss of life, damage and community panic.

Up until very recently, one of the primary challenges faced by technology-based intrusion detection systems dealt with the large number of nuisance alarms generated by these systems. Transit authorities have made significant investments in these systems only to have them turned-off or chronically ignored due to the excessively high percentage of nuisance alarms that render the systems, for all practical purposes, operationally useless. And having spent their budgeted funds, they are unable to replace these inadequate systems.

An emerging technology which enables these detection systems to be able to provide higher Probabilities of Detection [Pd] and lower Nuisance Alarm Rates [NAR]) involves the introduction of Deep Learning Neural Network (DLNN) software and the availability of affordable complementary hardware, integrated together to enable, in real-time, a much more operationally-effective system level of performance.

Sensors such as acoustic fiber sensors, video motion detection and radar are utilized for initial detections and real-time video—from appropriately located cameras along the tracks or even in the front of the train—is automatically processed by the combination of geo-spatial video analytics and DLNN software to effectively replicate what a human would do, which is to “verify” that the detection was an object (person, vehicle, animal, large obstacle) of concern and that it was, in fact, located on the tracks and, then, only if the object “classifies” as such, issue the appropriate alarm.

This practical use of DLNN technology can dramatically reduce the number of nuisance alarms generated from traditional detection only systems. Additionally, these systems can quickly provide imagery, as well as short looping videos to operators faced with making immediate and critically important decisions; namely, whether or not to stop a rapidly approaching train.

These enhanced detection with classification systems can be integrated with train control systems to automatically issue or provide a recommendation for a zero-speed control command. Of course, it’s always wise, time permitting, to have a human operator make the final decision on stopping the train. A further benefit of this new DLNN technology is that it can be added to many existing systems without having to abandon the investment already made in those detection systems.

By adding DLNN software to new and existing rail intrusion detections systems, the accuracy of these systems dramatically increases and the systems can finally deliver on their promise of helping to reduce the risk of trains colliding with people, vehicles and other large obstacles. Additionally, critical bridge, tunnel and other rail infrastructure can also be protected by accurately detecting and auto-verifying the presence and actions of those who would attempt to damage them, providing authorities with the credible and timely actionable intelligence required to take immediate action.


Larry Bowe has served as president of PureTech Systems Inc. since its inception in 2004. Prior to founding PureTech, Bowe was the vice president of Business Development for Verint Video Solutions and a director of Engineering at Honeywell Homes and Buildings Solutions.

About the Author

Larry Bowe | President, PureTech Systems

Larry Bowe has served as president of PureTech Systems Inc. since its inception in 2004. Prior to founding PureTech, Bowe was the vice president of Business Development for Verint Video Solutions and a director of Engineering at Honeywell Homes and Buildings Solutions. 

Bowe holds a B.S. in Computer Science from Arizona State University School of Engineering and an MBA from the University of Phoenix. Bowe completed a number of courses in e-commerce from Stanford University.