By incorporating smart tags on helmets one can adjust the local light conditions according to the safety prescriptions. There are hundreds of people on a site at the same time. The light has to react with low latency to be able to adjust correctly. In logistic halls the tags are attached to containers of which there are potentially hundreds of thousands. To be able to track those it is important that one can follow these "on the edge" while not having the computational capabilities to actually store the identifiable information (for example a MAC address) in its entirety. The assignment will address this in the following ways: 

• Study the methods for so-called approximate matching in network (variants of Bloom filters [1], Cuckoo filters, etc) 

• Study tradeoffs in network bandwidth required for communication versus local processing taking into account that neighbors see the same tags (it is distributed variant of above filters).  

• Design the asset tracking system taking also the non-functional aspects such as low-latency into account.  

• Simulate and analyze the overall behavior in a network simulator modeling also network loss. 

• Implement it in the firmware on actual devices. 

• Analyze the real-world behavior of a large scale system. 

In a nutshell, the challenge is that scanning devices can't make local decisions on tags in their vicinity, but need to implement a distributed algorithm taking into account information from their neighbors. That information should not hog the network though. 

References: 

[1] Bloom filters for data aggregation and discovery: a hierarchical clustering approach, P. Hebden, A.R. Pearce, 2005 International Conference on Intelligent Sensors, Sensor Networks and Information Processing, Melbourne, Australia 

Compensation: This internship is paid.