Adaptive Segmentation of Streaming Sensor Data on Edge Devices.

Area of Science:

• Signal Processing
• Data Compression
• Time Series Analysis

Background:

• Sensor data streams often exhibit twice-differentiable properties, requiring specialized segmentation methods.
• Existing algorithms may not efficiently handle real-time segmentation, smoothing, and compression simultaneously.

Purpose of the Study:

• To develop an adaptive streaming algorithm for segmenting twice-differentiable sensor data streams.
• To enable real-time simultaneous segmentation, smoothing, and compression of sensor data.

Main Methods:

• The algorithm employs a greedy look-ahead strategy.
• It is built upon the concept of a cubic splinelet for signal representation.
• Segmentation quality is evaluated by approximation accuracy and compression ratio.

Main Results:

• Achieved compression ratios ranging from 135 to 208, reducing stream sizes significantly.
• Maintained approximation errors comparable to state-of-the-art global algorithms.
• Demonstrated simultaneous real-time segmentation, smoothing, and compression.

Conclusions:

• The proposed algorithm effectively segments, smooths, and compresses twice-differentiable sensor data streams in real-time.
• It offers a practical solution for applications like IoT analytics and embedded time-series databases.
• The method balances high compression with low approximation error.