Entity reconciliation in a multi-camera network
Location traces are becoming fairly abundant with the introduction of various mobile devices such as smartphones, in-car navigation units, and video cameras. Each individual type of device generates different features about a mobile entity along with the location of that entity itself. For example, the smartphone can provide the motion (using accelerometer) of an individual, whereas a video camera can identify what type of clothing the person is wearing. A key challenge is to be able to fuse the data across different data sources and generate a unique view for each entity. This paper tackles a slice of this larger problem, which is to reconcile entities across a multi-camera network and a GPS trace from a smartphone and proposes a novel algorithm that can scale horizontally to adapt to new age distributed systems such as Apache Spark and IBM's InfoSphere Streams. We show through extensive experiments on a real-world dataset that our algorithm outperforms existing approaches and adapts to horizontally scalable distributed environments.