You mean you have two time-series data (Pushup vs noPushup) and want to separate them from each other? Also, the graph is not quite clear on the label. I need more descriptive qualitative information on the vertical x-axis.

Maybe.

I work with data that looks like this. I have measurements of resistance in a circuit with an insect wired into it. The insect's mouthparts touch the plant, and resistance is lowered. Resistance changes as the insect salivates, and as the mouthparts penetrate specific tissues in the plant. I can categorize at least 7 different patterns and I want to mark them and find when they happened and how long the insect remained in that behavior.

Look at Windaq waveform browser. There is a free version of the software.

http://www.dataq.com/products/windaq/#nested-tab-3

Now the problem is whether you can trick the software into reading your data file. This might be very easy, or not. You will not know until you try.

The software is not automatic. You have to go through the file using the software and mark the file. In this case I might use P (for pushup) and R (for rest). You mark the end of the activity with the letter that corresponds to what you have just finished. The computer calculated the duration of each activity and outputs a data file. You take that data file and do what you want. Windaq will also give you the measurement at a specific time. So I note in your graph that the rests become longer over time, the depth of the spike down diminishes over time, and the width of the spike down also increases over time. The analysis could become fairly complex.

Juehui Shi my mobile accelerometer generating data at 20 milli seconds. The X.axis shows the data point at the difference of 20 ms. The Y.axis is the actual values of the accelerometer. As I wrote earlier that the first 800 point on x.axis show nopushup, mean I have just started the application and put mobile into my pocket, and from 800 to 3000 there is pushups in my case its 10, and from 3000 to 3800 shows that I finished my push ups and take out the mobile from my pocket. Now the task is to Label these two categories accordingly. Question when we deploy the real application, then how we figure out that the incoming values are lie in pushups or nopushups? Either we will design a fixed moving window size lets say 500 points, then how could find the real count in this window, or please suggest some other way to do that? Bundles of thanks.

Ah, this is not a data analysis question it is just how do I detect a pushup in these data. A moving frame might work. It could also work to compare two data points separated by some time interval. In either case you will need a decision rule for when a pushup starts and ends. You could also just use a cut-off value. If the signal drops below -5 then you have a pushup. Increase counter by 1. If it rises above -5 then reset the detector and wait until the signal drops below -5 again.

You will have to experiment. What happens if you are very slow in your pushups? It takes me ten seconds to go down and back up (sort of a Tai Chi style pushup). Now what? What if they are done badly? Maybe this App will discourage the "pushup" where the arms barely flex. How about I simply collapse on the floor and then slowly push my way back up, then collapse (another bad strategy)? What does it look like if you are super good and push up off the floor and clap your hands? One handed pushups?

Ahmed, do you have raw data besides the graph? If so, you need to assign data point range (defined by you) to pick up those pushup data observations. SAS or Excel "if" function might work. Once you have two sets of pushup and nopush, you can then further your analysis.

Your graph is interesting though why is x-axis on the vertical?