# Need help with accelerometer data processing

• Shyam Krishnan
In summary: If you have a series of accelerometer measurements that are all oriented in different directions, you can use a linear regression to fit a Gaussian function to the data and then use the random noise as a noise predictor for future measurements. This will give you a good estimate of the orientation of the vehicle.
Shyam Krishnan
Hi,
am working on a vehicle tracking device, i am using LIS3Dh accelerometer to get the acceleration data.
i am using TM4c1231e6pz controller. I need to implement a harsh breaking alert, I am having difficulty in finding out which direction the vehicle is moving since the axis are not aligned with the vehicle. it can be fixed in any direction in the vehicle. I checked online I found methods to find the total acceleration only. can anyone tell me how to find the orientation of the vehicle and the direction in which its moving irrespective of how the device is fixed on the vehicle. I don't have a gyro sensor .

am desperate...

Total acceleration could work, assuming harsh breaking leads to larger accelerations than driving in a curve or accelerating forwards. Otherwise you'll need some calibration, assuming the orientation in the car is unknown but fixed. First get the direction of gravity - that is easy. Then you are just left with two dimensions. Traffic lights should work for that: the car has acceleration in some fixed direction (it stops), then absolutely no acceleration for some time, then acceleration in the opposite direction (it starts moving again). That pattern should be unique to traffic lights (and maybe stop and go traffic), and it gives you access to the forward/backward direction. Using multiple traffic light stops can improve the accuracy.

mfb said:
Total acceleration could work, assuming harsh breaking leads to larger accelerations than driving in a curve or accelerating forwards. Otherwise you'll need some calibration, assuming the orientation in the car is unknown but fixed. First get the direction of gravity - that is easy. Then you are just left with two dimensions. Traffic lights should work for that: the car has acceleration in some fixed direction (it stops), then absolutely no acceleration for some time, then acceleration in the opposite direction (it starts moving again). That pattern should be unique to traffic lights (and maybe stop and go traffic), and it gives you access to the forward/backward direction. Using multiple traffic light stops can improve the accuracy.

what you suggested could work, my doubt is ,suppose the device is fixed in such a way that when the vehicle moves forward it is at angle 45 degree with the x and y axis, then both x and y-axis would have components of acceleration ,so from both x-axis and y-axis data how can I calculate the acceleration ?

With a coordinate transformation.

mfb said:
With a coordinate transformation.
I know, Its possible when we know the angle, but in my case when I give my product to a client , I have no idea how they are going to fix it (they might fix it upside down sidewise or just leave it hanging from the connector etc..) so the angle is also unknown... that's my real issue

That's the point of the traffic light stops to calibrate it (=figure out how it is oriented in the car).

If it is hanging freely, or changing its orientation frequently, forget it.

mfb said:
That's the point of the traffic light stops to calibrate it (=figure out how it is oriented in the car).

If it is hanging freely, or changing its orientation frequently, forget it.

ok. Anyway thanks a lot. some of your ideas are really good , I will work on it. and try to calibrate it using the traffic light idea

I have a couple thoughts. I process accelerometer data [almost] daily so I'm quite familiar with working with them. Be sure to look up whether your accelerometer excludes the effects of gravity (many do).

1.) First thing I would do is integrate the acceleration time history (look up trapezoidal numerical integration) to get the velocity components. Then, do this again to get displacement. (One word of caution, you may need to high pass filter the data before each integration at approximately 5 Hz to remove the DC component which can make the Vel. TH and Disp. TH incorrect). The vehicle is going to displace the most in the direction it is traveling and the total velocity will be the greatest in the direction of travel as well. I would plot all of the velocity time histories together and the displacement time histories together. Since it's not oriented with the vehicle axis, it will be at minimum a combination of the 2 directions (or possibly all 3) to figure out the orientation. If you do the entire time segment, you should see where braking begins, acceleration starts, etc. If you see that 2 velocities and/or displacements are near the same and are the highest, then those 2 are in the direction of travel 45° from the axis of travel and the other direction is orthogonal.

2.) Additionally, road vibration is mostly Gaussian random vibration (although some sources have shown vehicle random vibration is actually more random than random vibration...i.e. higher Kurtosis value). So consider that you will have random vibration, vehicle acceleration (and deceleration and centripetal motion), and shocks (and also noise) possibly all in one single time segment measurement. This is not just some object accelerating in a straight line in a vacuum with no outside influence. These can all be converted to power spectra, but it may not be as beneficial... but things such as velocity Power Spectral Density (PSD) and displacement PSD can be computed--something else to consider calculating. So in other words, don't be surprised if the results don't come out to exactly what you anticipate. Filtering of the data may be required to get what you want.

## 1. How do I collect accelerometer data?

To collect accelerometer data, you will need an accelerometer sensor and a device capable of recording and storing the data. The sensor can be attached to the device or worn by the person/subject being measured. The data can then be downloaded onto a computer for processing.

## 2. What is the best method for processing accelerometer data?

The best method for processing accelerometer data depends on the specific goal of the study or experiment. Some common methods include filtering, integration, and statistical analysis. It is important to carefully consider the research question and determine the appropriate processing techniques.

## 3. How do I deal with noise in accelerometer data?

Noise in accelerometer data can be caused by external factors such as vibrations or movements that are not related to the study. To deal with noise, you can use filters to remove unwanted frequencies or perform signal processing techniques such as wavelet analysis. It is also important to ensure that the accelerometer is properly calibrated before data collection.

## 4. What are some common challenges in accelerometer data processing?

Some common challenges in accelerometer data processing include dealing with missing or erroneous data, deciding on the appropriate sampling rate, and selecting the right processing techniques for the specific research question. It is also important to consider any potential biases in the data and address them during processing.

## 5. How can I interpret accelerometer data?

The interpretation of accelerometer data depends on the specific research question and the processing techniques used. Typically, the data is analyzed to identify patterns or trends, and statistical methods are used to determine the significance of the results. It is important to carefully consider the limitations and assumptions of the data and the processing techniques when interpreting the results.

• Mechanics
Replies
11
Views
23K
• General Engineering
Replies
9
Views
8K
• Electrical Engineering
Replies
2
Views
2K
• Mechanics
Replies
3
Views
1K
• Introductory Physics Homework Help
Replies
17
Views
2K
• Mechanics
Replies
2
Views
2K
• Mechanics
Replies
4
Views
15K
• Programming and Computer Science
Replies
4
Views
4K
• Programming and Computer Science
Replies
8
Views
2K
• Mechanical Engineering
Replies
1
Views
1K