Detect. Report. Deter

A FUTURE SYSTEM TO PREVENT SEXUAL ASSAULT ON THE TUBE
USER-FOCUSED DESIGN • FORESIGHT PRACTICES • PSS DESIGN • PROJECT MANAGEMENT

Brief

Utilise futures thinking, analysing social, environmental, technological and economic trends to predict a global issue in the year 2042, then design an intervention utilising forecasted technologies.

Outcome

Sexual assault is extremely damaging on a personal and societal level, and has been shown to worsen as urban spaces become more crowded. Environmental pressure, reduced use of private transport, and a rising population have led to an increased use of public transport in London. However, supporting TfL infrastructure was not given sufficient time to develop due to a lack of significant government intervention, resulting in overcrowding, making it easy to get away with contact-based sexual assault.

Detect. Report. Deter. uses CCTV on tube carriages to monitor for signs known to be caused by sexual assault using EVM and emotion recognition. This triggers collection of Wi-Fi data from passengers’ IoT devices. A 3D model is then created to capture the poses of passengers, even parts obscured by traditional CCTV. Sexual assaults undergo a three-layer screening process with two AI algorithms and a human to ensure a low false positive rate. The victim is notified on their device and they can file a report, knowing the evidence is there.

Future Contextual Study
‍Analysis of signals, drivers and trends in a literature review led to the development of the future scenario, identifying why the issue cannot be solved now and the primary problems for the user - including non-functioning CCTV, feelings of responsibility from the victim and it not being safe to intervene. Technological enablers close to now, such as 5G rollout and edge computing, and ones further on, such as WiFi sensing and long battery life were identified, and we engaged in repeated discussion with the CTO of TfL and a professor of design against crime to further validate the predictions and the technology integration capabilities.

Recognising Behaviour
Anonymous accounts were taken from online and personal experiences, which were analysed to show how emotional responses are tied to physical ones, and used to differentiate sexual assault from other emotions, and prevent the EVM system triggering incorrectly.
EVM is a technique for amplifying subtle motions in video so that they are more noticeable. Using a combination of micro gestures, eye tracking and heart rate monitoring, it is possible to detect specific emotions that indicate sexual assault. Applying EVM to CCTV footage then running emotion detection algorithms increases detection accuracy and can be run locally on CCTV cameras through edge computing. A prototype was built in Python. Despite the limitations of today’s technologies, EVM was found to be a powerful tool for identifying expressions.

Person Identification
A large issue is that on a crowded carriage, CCTV cannot ‘see’ as areas are blocked. Wi-Fi sensing detects and follows movements or gestures utilising signals from IoT devices, for example phones. Wi-Fi sensing data can be converted into 3D skeleton diagrams of passenger’s positions through the use of a neural network, even if there are obstructions.

A prototype was created using two ESP32 microcontrollers. We were able to obtain models in a 20 metre range (sufficient for the London Underground where carriages are between 15 and 17 metres in length) and up to 4.6 cm accuracy. In 20 years this technology would be able to reliably detect contact.

Reporting
When a crime is detected and the system triggered, CCTV captures the appearance of the victim and perpetrator. Feature decomposition builds a picture using personal features to be able to identify the person even if they change their appearance. When the person touches out, cameras at the barriers scan them looking for a match from the incident.

Potential victims receive a notification on their lockscreen through the TfL app. Unlike a traditional notification, it is interactive so that a report can be made without having to unlock the device. In-app, users can opt in to receiving these notifications - but people who do not wish to, out of privacy concerns or due to previous experiences, do not have to. The outcome is shown in the flow diagram.

Privacy/Data
‍Privacy is at the core of the system. Data are only gathered when the system is activated, are stored in a local buffer, and data flow is one way - personal data cannot be identified from the stored information, only becoming accessible with opt-in consent or in an investigation. Human validation is always used to verify an assault has occurred before notifying the victim.

Despite the complex technology working in the background, the process is very simple for the user. For all travellers, the journey process remains unchanged, and for victims, the system mitigates the pitfalls of the current system, integrating seamlessly.

Implementation Roadmap
To make this a reality, actions from various bodies were identified. No one will do something just ‘to be nice’, so benefits of engaging in the system was established to encourage uptake.

An implementation roadmap was created, with future development opportunities. Barriers to implementation and potential pitfalls were outlined, and solutions identified so implementation can be as smooth as possible.