11.1 Intelligent CCTV combined with sound warnings

Description

What does this measure refer to and what is its objective?

CCTV systems can be used to monitor the entrance of unauthorised persons to the surveillance area. Intelligent or smart CCTV uses algorithms able to recognize situations in which people move into an area that they should not enter, or behave in a way that places them at risk. When such persons are detected, they are automatically given a sound warning (alarm noise or spoken message) by loudspeakers. The main goal is influencing the person to modify their behaviour and move to a safe place. Another goal is to facilitate the surveillance process and to record relevant footage which can be used later on, for example in risk assessment.

Recommendations

Best practice and lessons learned

  • When used at stations, cover platform ends. One common procedure is the installation of a camera at the tail end of the platform with the image visible to staff at both the platform level and in the control room.
  • The sensor should be “trained” to recognize only the relevant “targets” and “patterns”.
  • The sensor needs to be able to react only to persons who are in its range.
  • Nowadays there are many different types of cameras available and each type may be more suitable for a particular context (day, night, low light, distance range, etc.). For example IR cameras are better for night detection. In some situations systems with dual cameras may be needed (e.g. normal and thermal camera).
  • For trespass, the measure requires a follow up to remain effective. It should also be combined with measures such as human surveillance, prosecution based on pictures collected with the CCTV systems, local intelligence, etc.
  • For suicidal people, the spoken message may have the effect of interrupting the "impulsive mood" of the person.
  • For suicide prevention, a framework has been produced with five main classes (display of emotion, appearance, posture/movements, activities and interactions) and associated sub-classes (Ryan, 2018). Commentary has been provided on factors that influence identification of suspicious behaviours, how to distinguish these from normal behaviours and the circumstances that inhibit timely reactions to the behaviour amidst the complexity of the operational railway.

Warning points

Expected difficulties and issues you should pay attention to

  • It is not known how a person that contemplates suicide would react to an alarm sound.
  • Be aware that sound warnings may lead to noise pollution which can cause acceptance risks with neighbours and nature conservation organisations. You may encounter aversion and resistance against the system from the people living in the direct environment. May not to be used in rural nature areas because of disturbance of fauna. Communicate before installing. Even after installation, station announcements may not be permitted during some time intervals (e.g., between 23:00 and 07:00) due to issues with noise pollution affecting neighbours. For further details on how to reduce noise pollution you can check http://www.advanced-noise-solutions.co.uk/cs-noise-pollution.html
  • Children could play at triggering the sound system as a game.
  • Intelligent systems may fail to detect relevant situations (false negative errors) or detect irrelevant patterns as relevant (false positive errors). While false negatives are totally unacceptable, when purchasing a system make sure the false positive proportion is also as low as possible.

Observations

Other points that you should not forget

  • The intelligence of the system may largely vary. New smart CCTV systems can be trained to identify abnormal patterns of behaviour. “Non-intelligent” CCTV systems send information to the station staff for further actions. Implementation depends on the intelligence of the technical system (active or inactive), i.e. whether the system is actively monitored by the staff.
    • If the system is intelligent and makes its own correct assessments of the situation, the measure has no impact on other measures.
    • If the system needs to be monitored actively e.g. by traffic control it has an impact on the skills of staff. If this is the case you should make sure the staff are properly trained. If the images are analysed afterwards, qualified staff with other qualities are needed.
  • Although in the context of RESTRAIL this type of detection system is recommended against suicide, most publications and field tests already conducted consider these devices as anti-trespass technologies.
  • In Sweden, the cost for the installation of video cameras combined with motion detectors covering the track area where trespassing occurred frequently was €700,000 (6 million SEK). Contact person: Helena Rådbo.
  • The use of drones with thermal imaging can be considered to address trespass. With a drone it can be easier to search and detect a trespasser. An example here. Drones may reduce search time and allow trains to continue to operate, as well as expedite the safety in the area, but operating drones requires compliance with national UAV regulations.
  • With the development of video analytics (VA) centralised artificial intelligence (AI) solutions can detect and classify hazards (e.g. Zhang et al., 2022): information is collected for real-time remote processing to calculate the risk profile probability of an intrusion event based on the location, classification, direction of movement and proximity to the rail danger zone as pre-determined in scoping. Such a solution would reacquire less physical barriers and can automate the administrative alarm management process for network controllers to ultimately reduce the likelihood of human error, and provide an interface to the train control system to stop trains under agreed scenarios (source).

Study results

Data or other evidence supporting the measure's effectiveness

  • Overall decrease in trespassing behaviour was experienced over time: 46 trespassing events for the first year, 18 for the second year, and 38 for the third year. 60% drop in trespass from the first to the second year and just a 17% drop in the third year compared to the first. An increase from the second to the third (DaSilva et al., 2006).
  • Linking CCTV with public address-system-announcements may be effective in deterring children but less effective for adults (RSSB, 2005).
  • The system can also be used to deter vandalism (RSSB, 2005).
  • A detection system composed of CCTV combined with sound warnings was tested by VTT in Finland as part of RESTRAIL pilot tests conducted in 2014. In that pilot, a significant reduction in the daily number of trespassing was obtained (Kallberg & Silla, 2016). Also read a press article here.
  • A system consisting of a camera, IR spotlight and loudspeaker was tested by INFRABEL in Belgium at a railway tunnel trespass hotspot situated in a densely populated area. In terms of accuracy of the system, there were 30% false alarms (mainly train detections). In terms of safety effects, there was a -80% trespass reduction in 2018 compared to 2016/2017 (reference period when the system was not in place). The evaluation of the system continues.
  • Speakers in combination with video detection cameras will be evaluated by ProRail in The Netherlands in a 6 to 12 months pilot study.
  • Using CCTV footage as data source provides valuable insight into relevant situational conditions in which suicides take place, which can be useful to inform prevention strategies, particularly information about behavior and place combined (Ceccato et al., 2021)
  • Artificial Intelligence (AI)-aided framework for the automatic detection of trespassing events used deep learning-based tool to automatically detect trespassing events, differentiate types of violators, generate video clips, and document basic information of the trespassing events into one dataset (Zhang et al., 2022).

last update: 2022-07-29