Real-Time Classification of Streaming Sensor Data

The last decade has seen a huge interest in classification of time series. Most of this work assumes that the data resides in main memory and is processed offline. However, recent advances in sensor technologies require resource-efficient algorithms that can be implemented directly on the sensors as real-time algorithms. We show how a recently introduced framework for time series classification, time series bitmaps, can be implemented as efficient classifiers which can be updated in constant time and space in the face of very high data arrival rates. We describe results from a case study of an important entomological problem, and further demonstrate the generality of our ideas with an example from robotics.