Hello,
I have some GPS trajectories that include gps locations recorded on variable frequencies (every 1 min ,5 min , 20 min,etc.)using a geographic projection WGS84. In my work, i need a fine-grained GPS collection (e.g. 30 sec) to be able to recognize accurately users' activities grom their GPS traces.
I thought then to generate the missing GPS points between two known points (for instance, when the points frequency is every 5 minutes, i need to generate the 8 missing points that should be recorded the frequency was every 30 second). Note that i have to estimate the speed of the missing GPS points in my interpolation . Could you suggest some mathematical estimation or an existing tool that can answer my issue?
Best.
This appears to be basically a resampling/upsampling problem... This is also what is done by tracking algorithms... May I suggest studying "tracking" and "trackers"??? These generally involve Kalman filters to essentially smooth the motion, allowing estimation of the inbetween samples...
Hello Mehdi:
The interpolation depends on the nature of the data, if the known points give a quite stable trend or pattern, it may be quite promising to use a function with certain type of form. If the interpolation is applied to unknown patterns or trends, model selection can be very sensitive, because the interpolation may incorrectly interpret the underlying pattern. Usually, if one is uncertain about the underlying pattern, he/she may try 1) non-parametric estimation, for example, kernel method, and 2) combine multiple non-parametric function to decode the latent patterns.
Below is a link with information for spatial interpolation methods, these are the most common approaches we are using in our research:
http://www.gaussianprocess.org/
If these don't help, sorry for the irrelevant information.
Good luck,
Jiong
the important point to note in Jan Wong's useful reply, by the way, was the importance of underlying patterns and their supposition. IT is one thing if the pattern is clear and another if it is not. An EMpirical Baysean solution may serve to smooth where a strong a priori is available and you need to add some points. But where a good a priori model is absent it's a different story. In a program like ArcGIS there are kerneling functions whose algorithms may help but there you need to be careful about things like the method being used, size of the kernel, etc.
Mehdi,
I would use the methods included in the attached paper for missing data imputation.
They are so easy to implement in R to impute various data behaviors.
I can guide you about that if you prefer to use R.
Best,
Conference Paper Comparative Statistical Algorithms for Imputation of Missing...
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics
- Geographic Information System