In some situations, human safety can be improved by bestowing simple vision capability upon machines, giving the ability to detect humans and movement. Yet, previous attempts to do so have proven unreliable and ambiguous, partly because of the complexity of visual analysis. The EU-funded D-SENS (Depth sensing systems for people safety) project approached the subject from a new direction. The 11-member consortium aimed to achieve safety applications of machine vision by combining 2D colour information with 3D depth data. Using two such different channels to represent visual information helps remove ambiguity. Other goals included establishing a relevant research team and marketing the results. The undertaking achieved solutions involving algorithms for detecting and tracking people in public spaces, including retail and road environments. The systems were also able to detect left items and intrusion. The solutions combined high-quality data, in terms of depth and intensity, with modern vision algorithms. Results yielded a robust visual safety system. The accuracy came from reliable representations of scene geometry, which remained unchanged following variation in light levels or colour. The machines recognised several key concepts adequate for practical use, including: humans, left items, intruders, scene models, motion patterns and activity. The prototype system was successfully demonstrated via five use cases in the fields of smart buildings, assisted living and security. Researchers also created and validated a Common Framework, which shares functionality among prototypes. Demonstration sessions illustrated the strengths and weaknesses of the respective version, paving the way for future commercialisation. The D-SENS project means new ways for machines to visually detect and track humans. The development should improve safety, while also representing commercial opportunity for European business.
Safety, algorithms, visual data, depth sensing, tracking