An article prepared by SquareDev (Partner of the LAW-GAME Consortium)
Lessons learned from our involvement in the LAW-GAME project
LAW-GAME implies a Digital Gamification Approach for effective experiential training and prediction of suspicious actions. SquareDev (SQD) is an experienced startup in the computer vision area, with an extended experience as a technical partner in EU projects, handling tasks that include state-of-the-art Artificial Intelligence implementations, as well as Deep Learning methodologies.
The scope of LAW-GAME is to elevate experiential training for police officers through gamification technologies. The main common goal is to train Law Enforcement Agencies (LEAs) in developing advanced skills and competencies for intelligence crime analysis and illegal acts prediction. LAW-GAME will develop and design a training system based on serious games, virtual reality and Artificial Intelligence assisted procedures. The project will develop an advanced learning experience to train LEAs and measure their proficiency in conducting a forensic examination, effective questioning and recognizing and mitigating potential attacks.
The project’s end-users will engage in a learning experience to develop key competencies needed for successfully operating within diverse and distributed teams; an important skill when collaborating in various cross-organisational and international cooperation situations. The project partners will collaborate to create real-life simulations utilising Artificial Intelligence and virtual reality for practitioners in four focus areas including crime scene investigation, interrogation and negotiation, terrorism prediction, vehicle dynamics and car accident analysis. [1]
What is the role of SQD in Law-Game?
Within the LAW-GAME project, the tasks, closely related to Computer Vision techniques, are without any doubt, challenging. The role of Squaredev within the project is twofold. Firstly, the Computer Vision team provides an image annotation module as a microservice to detect suspicious items from accident and crime scene pictures, and secondly builds another hostile pose detection service, to recognize the exact pose of a human.
In collaboration with the University of Salamanca (USAL) and the Centre for Research and Technology Hellas (CERTH), SQD tries to implement a series of algorithms to identify suspicious objects in crime and accident scene images using Computer Vision techniques for the 3D reconstruction of the scenes.
Having the task, from the part of SQD, of detecting the position of various objects in images of crime and accident scenes some challenges were raised. The most significant of them was the nature of the data. As the list of objects included several objects of different natures for about 90 categories like cartridges, vehicles, skid marks, scene evidence etc. the open-source data options were limited. The process needed manual gathering of the dataset and manual annotation. In this procedure, the LabelMe app [2] proved to be very useful. Having tackled the major challenge, SQD showed significant progress in detecting suspicious objects using state-of-the-art Computer Vision models like Detectron2 [3]. The results were impressive, especially for pretty challenging images like the skid marks and multiple types of vehicles within the same images, for car accident scene reconstruction.
A sample with open data sources is provided below.
SQD’s mission is to use technology for a better tomorrow for everyone. We research AI so we can enable CTOs around Europe to build innovative solutions for their users and customers.
Our role in the LAW-GAME project will further enable the company to develop and work towards reaching its mission.
For more information visit the LAW-GAME website at: https://lawgame-project.eu/
Stay tuned on the project’s latest updates by following our social media channels at:
LinkedIn: https://www.linkedin.com/company/law-game/
Twitter: https://twitter.com/LawGame_EUH2020
References
[1] “Overview of Law-Game”, Retrieved November 2022, from https://lawgame-project.eu/
[2] LabelMe repo on GitHub: https://github.com/wkentaro/labelme.
[3] Detectron2 repo on GitHub: https://github.com/facebookresearch/detectron2 .