This paper discusses the use of a sonar sensor that can be placed on a container to predict the volume of content inside the container. They send an active signal through the container and then measure the impulse response at other side until after a second. They call this acoustic probing. They do this for N-levels of contents and then use Support Vector Machines(SVM) to predict the level of the food inside. Since the volume might not match exactly into N levels, they then use Support Vector Regression(SVR) to transform those values into continuous levels instead of discreete levels. They continue to do this for many household containers for pantry. The solution provides robust method of non-deformable containers but the accuracy is very bad for deformable containers like chips and milk packets.
I found the application this paper quite interesting. While I think a general-purpose volume prediction has a long way to go, using it for a specific product in any industrial or commercial setting seems very much achievable.
- The paper does not discuss how the confusion matrix might change as the container gets old and its wall size changes. For precision, that needs to be handled as well.
- How to use Support Vector Machines(SVM) for prediction problems?
- How to use Support Vector Regression(SVR) for regression problems?