Advertisment

Multi-sensor fusion leads to better accuracy, precision: Summit of Things 2024

We need to be able to do object identification and association, registration maps measurement to common coordinate system, and combine data uncertainties. 

author-image
Pradeep Chakraborty
New Update
gerra
Listen to this article
0.75x 1x 1.5x
00:00 / 00:00

Summit of Things 2024 was held today in the United States.

Advertisment

Johannes Beekman, CEO, IoT Marketing, talked about advancements in edge AI. He said that edge AI has future such as next-gen use cases, integration with AIoT, neuromorphic computing and spiking neural networks, and ethical and regulatory considerations. Let us embrace edge AI for future innovation.

Scott Gerard, Knowledge Reactor, gave the keynote on multi-sensor fusion. Data fusion is the integration of data from multiple sensors to produce a unified picture. It is uni-modal, using multiple sensors of same type, and multi-modal, using multiple sensors of different types. These are widely used for surveillance, autonomous vehicles, robotics, etc.

Multi-sensor fusion leads to better accuracy and precision, better redundancy and reliability, and better discrimination. We need to be able to do object identification and association, registration maps measurement to common coordinate system, and combine data uncertainties. 

Advertisment

Let say we have to track moving target use case. You can use sensor data and fuse them together. We use Kalman filer and unscented filter (UF). We may also have to do particle filtering. The goal is to return an X-Y coordinate of a target. 

Another interesting case is in precision agriculture. We can monitor plant health, and minimize water, fertilizer, and pesticides. The goal is to figure out how healthy the plants are. We apply only to the plants that need them. 

Another case is sleep quality use case. We needed to have occupant's recommendations, and control HVAC. We followed the process of Bayesian networks and neural networks. We did not have very small training data sets. 

Advertisment

We also use knowledge graph as the sensor. We can search to appropriate models in the KG. We can extract data from neighborhood KG nodes and edges. We can then fuse like any other data. 

Small training datasets have rare anomalies. We have simple models with few parameters to learn. We can also do data augmentation, and have synthetic data generation. Data fusion is an evolving field at the intersection of technology and math.

iot-sensors
Advertisment