Which metrics do you use to compare sensor models and sensors for localization?

I do not understand the question

It is normal that the sensors respond differently based on differences in their environment !??

e.g. a laser will give a very good result is you want to match it with an other laser scan in a typical office environment, but in a long corridor it is very difficult to detect the location in that corridor since all scans will be the same. WiFi fingerprinting at the other hand will give you only a rough estimate of a position in a room but will be able to identify the area where you are when you are in a corridor.

If I understand correctly, you have two modes of detection (a laser and a Wifi fingerprint technique, I do not know this technique)

it is normal that two modes of investigation give you two different kind of results

you need to mathematically analyze these two signals that in order to extract relevant information.

Does this answer your question?

No; Maarten and I agree that the signals must be mathematically analyzed, yet we are looking for established metrics to do so. We find one possibility to be information theory as used in [1], where simple sonar range sensors were compared with a biomimetic sonar system.

Information theory seems to be a very suitable way to compare distinct sensors. Other approaches usually seem to be qualitative, in the sense that the resulting localization performance is compared rather than the sensors themselves. What are the metrics used when comparing possible sensors for a localization algorithm?

[1] J. Steckel and H. Peremans, “Biomimetic sonar for biomimetic SLAM,” in Sensors, 2012 IEEE, 2012, pp. 1–4.

It is normal that you have a different response from your sensors. e.g you can use the same temperature sensor both in room and in a reactor. The response of each one depends of the overall system, the behaviour of your system and its modelling. So, you can't have a same result.

I think there is a misunderstanding here. We all agree that the result will be different. We are just looking for generic metrics which can be used to compared the information a sensor provides to an overall fusion of sensors.

True, we can not expect the same result for any sensor, especially not when comparing sensors as different as a laser range finder and Wi-Fi measurements. However, this exactly touches our point of interest: how do we compare these results?

From an information theory approach: which sensor provides the most information at what point? But then, is information theory the only or the proper approach?

May be this book chapter can help you :

Hi Lala very nice chapter but I do not how it related to e.g. the information a laser scan gives compared to e.g. the localisation of passive RFID labels on the floor or WiFi fingerprinting.

Hi Maarten, I suggest you throw a look at this paper. The authors propose a multiple-metric learning algorithm to learn jointly a set of optimal homogeneous / heterogeneous metrics in order to fuse the data collected from multiple sensors for classification :

Dear Lala,

This is an interesting paper!

The problem is also well known to the EVARILOS project, which tries to propose a universal benchmark for WSN localization.