MIDS Capstone Project Summer 2017

Fink Security


Fink Security is an AI based security platform that utilizes machine learning at the levels of "smart sensor" nodes and within the cloud in order to remove noisey data and empower communities to contribute to their own secutiry.

Newer security systems have very little innovation over previous generations. Basic entry detection is maintained, and innovation has been limited to tacking on video which streams indescriminantly to a central storage. Unfortunately false alarms are still occuring around 98% in some areas. Due to this frequency of false alarms and the high cost of analysing indescriminant data, police departments are unable to respond effectively and are opting either to charge for false alarms or not respond at all.


Fink Security takes the approach of analyzing data at the nodes and in the cloud, drastically reducing extraneous data and allowing for specific knowledge of entities and events. Our network of homes all contribute observations of what is going on in the area around them, and our AI system consolidates and analyses these pieces to create a larger understanding of events and the people involved. Faces, licensplates, and mobile device ID's are identified and stored along with times, durrations, etc. In the event of a crime, the system can reach out to neighboring homes to find answers that the victimized home can't provide. For instance, if a person wearing a mask were to break into a home, we can track that person back to a time before they were wearing the mask even if they were blocks away. Associations can be assertained when a suspect co-occupies a car, or is seen interacting with another person. This information, released by the customers who acquired it, is then presented to the police in the form of an evidentiary package. Durring an alarm, our system can assess a threat rating and determine if an actual crime is taking place or if it is a false alarm. This validation allows police to reduce their false alarm frequency and dedicate resources where they are needed most.


Nodes: "Smart Sensors" are implamented on raspberry pi units with infrared cameras. In production these would be replaced by proprietary units. These smart sensors collect, isolate, and report faces, license plates, and mac addresses. 

Cloud: Streams of acquired records are accumulated and processed into various storage structures. Faces are cleaned and processed via CNN into a 100 dimension vector, stored in a cube field in postgresql so that a cosin distance is indexed. License plates and mac addresses are also stored in postgresql with original images stored in S3 for evidenciary purposes. 


More Information

Last updated:

August 28, 2017