What is data fusion?
Data fusion is the method of fusing data to create new data. Fusion means that there is a combination of two (or more) different pieces of data to create a new piece of data (fused data). It is useful when you need to create new, meaningful data from previous data. It has five primary uses: (1) pre-processing data, (2) object assessment, (3) situation assessment, (4) impact assessment and (5) decision making. There is one question you can use to identify whether a module is a data fusion method: Is it creating a new data set with the previous sets of data? If the answer is yes, then congrats it’s a fusion method.
What is data fusion not?
Anything that is not combining two different pieces of data. One example is equations, like y=2*x. This is just calculating a value using known constant. Another example would be any module using mathematical models that are static and not updating based on current data would not be considered a fusion method.
What are some uses for data fusion?
Data fusion methods have a variety of uses, from preprocessing data before use, weighing the impact of outcomes and making a decision. Data fusion is used in multiple areas in engineering, such as feedback loops for physical systems, military systems and cutting edge technology. Currently, there is a huge popularity in deep learning that focuses on applications such as facial recognition (CNN, fast RNN), object segmentation and image reconstruction (GANs). This area encompasses a vast amount of data fusion algorithms that has so much potential to push us into the generation of technology. Namely, one area that will benefit is multimodal fusion, which is fusion of different modalities. There research on this type of fusion and its implementation in robotics to make robots aware of their surroundings in multitude of ways.
Other big areas that are implementing data fusion methods are autonomous vehicles and other motorized machines that will be interacting in a human environment. Autonomous vehicles will be very interesting to observe grow for data fusion, as there are a variety of sensor data that will need to be fused and will need to be able to work in real time. There are common fusion methods that are known to work, such as Kalman filter or Bayesian methods, but will the industry push beyond these methods and find something faster and more robust with all the different sources of data?
I hope you enjoyed this small post on defining data fusion. Check back on Monday for new posts!
Comments
Post a Comment