terrain classification using accelerometer

[Hari419]

I would like to perform terrain classification using tri axial accelerometer.I have connected my accelerometer sensor to PXI 4462 module.I have accquired data using DAQmx.Please help me out with Signal Processing of sensor data.

[ogk.nz]

This is all very vague,do you know what it is that you want to use to classify a terrain type from your accelerometer data?

How are you handling the accelerometer that it is gathering data about the terrain?

Can you read the data from the accelerometer?

Have you got any progress that you can show us how you have attempted this so far?

Generally the more information you give us and the more specific you are the more likely we are to be able to find solution to your problem.

[pjr1121]

The simplest classification settings would be based on amplitude thresholds.  Of course, the amplitude of the signal you receive will depend on both the terrain and the speed of the vehicle.  (Hitting a bump at higher speeds will register a higher acceleration).  Just drive over different types of terrain recording the data.  This will give you a basis for comparison.

[Hari419]

I have obtained some variations in Intensity of the accelerometer signals. But I need some statistical data of signals from different terrains such as Sand,Pebbles,Stones,Grass etc.So please help me out with some features

[RavensFan]

I doubt anyone here would have information like that.

You are going to have to do your own research.  Maybe even conduct your own experiments.

[johnsold]

The kinds of acceleration signatures you get from different surfaces will vary significantly depending on the dynamic mechanical structure of the vehicle suspension (tires and wheels, springs, shock absorbers, ...) and the speed at which it moves across the surfaces. The suspension will act like a filter to change the amplitude, frequency, and phase response of the signal. 

Are there standards in the field/industry to define these parameters or do you need to create your own?

Lynn

[Hari419]

Guys I have tried some stastical data in time domain such as Mean,Median,Mode,Standard Deviation,Variance and Kurtosis.But those features do not show repeated and unique value while measuring acceleration signals.So can you suggest me some data which give unique values while measuring.

[johnsold]

That can be a very difficult task.

<Sea Story> I once worked on a project to classify gear and bearing noises in a machine. We ran tests on about 100 machines. We had a list of which machines were believed to have which kinds of problems.  We looked at both time domain and frequency domain analyses. We made lots of graphs. We finally decided, somewhat arbitrarily, that this group of machines had similar results in one frequency band and that most of them had been classified as having gear problems. We created groups for each type of fault.  In every group we had some machines that did not "fit" the classification.

After putting the tester on the production line for a week or two, we made some adjustments in the classification paramters. 

Six month later the production manager called insisting that our test had gone out of calibration. Why? Because the failure rate had dropped from 50% to 10%. We took the tester back to the lab and checked the calibration. It was within a fraction of a dB of the original settings. What had happened? The production crew had started getting good feedback from our tester: If it said there was a bearing problem, there always was a bearing problem. The orginal testing method was very subjective and frequently wrong. Once the people building the devices began getting reliable information about the problems, they started building the devices better the first time. <End Sea Story>

My point is that when dealing with data which has a very high degree of variability such as sand, stones, and grass, you may not get simple, clean discrimination between the groups. 

In addition to the time domain measures, I would look at the frequency spectra. I would expect the sand particles to generate higher frequencies than stones. 

Lynn