QRS detection algorithm would be best.

els u can go for simple correlation or subtraction method using templet matching 

QRS, exactly. Why re-invent the wheel? Good algorithms already established (e.g. Pan-Tompkins).

Most software inspired "peak" detection software does not discriminate well without strict limits placed on the expected range of amplitudes and peak widths.

BTW: Use an a/d converter with high resolution and low noise.

Thank you guys for your answers.

I am not extracting data but using some recorded data.

There are many QRS detection algorithm like pan Tompkins, Hilbert transform etc, so which one is the better one considering i am doing my analysis on MATLAB.

Pan-Tompkins is an R-peak detection algorithm which works very well. The Hilbert transform, like the Fourier transform or wavelet transform is not an algorithm to detect R-peak (or QRS complexes), but a method which can be applied to pronounce the characteristic features of e.g. the QRS complex to improve final detection.

Some years ago I developed an algorithm for R-peak detection based on the continuous wavelet transform, which works well in settings with low to medium amounts of noise and artifacts with low computational load ("Low-complexity R-peak detection for ambulatory fetal monitoring") which compares favorably to Pan-Tompkins and Li. et al. ("Detection of ECG Characteristic Points Using Wavelet Transforms"), which at the time had the best published performance on the MIT-BIH database. More recently this publication ("QRS Detection Algorithm for Telehealth Electrocardiogram Recordings") seems to achieve very good results, although I have not tried it yet.

Which algorithm is best depends on the exact goal of R-peak detection and the environment in which the ECG has been recorded, i.e. in a clinical setting while the patient is still or unsupervised in an unknown environment with possibly a lot of movement artifacts. A big advantage of the Pan-Tompkins algorithm is that it is freely available. Furthermore, you have the goal of calculating the HRV, which means you want to obtain sample (or sub-sample) accurate detection of R-peak locations AND know if the detected peak is 'sure' to be an R-peak or might be either a noise spike or arrhythmic peak. A method which works well for me is to find extraneous peak by comparing consecutive RR intervals and marking all R-peaks which have resulted in an RR-interval which differs > e.g. 20% from its neighbors as a miss-detection. Then be sure to only use the clean RR-intervals in your HRV calculations, and only if no more than a small fraction of the R-peaks were determined as being artifacted.

 There are many, many methods of doing this. There isn't a 'best' one because the different methods have various strengths and weaknesses, and that depends heavily on the nature of your ECG.

I had a failure recently of some normal methods mentioned above, for instance, where the changes in peak height and baseline wander (it was an ambulatory recording from a child) totally defeated the algorithm. 

As any RR series you calculate should be visually inspected vs the ECG record in the first place, the usual concern - instead of finding the 'best' method - is finding a method that will get you close enough to check any errors, ectopy, etc. without over- or under-detecting.

peakdet.m is very simple, it just selects a value identified as local max above a stated threshold. It will work reasonably well with clean data if your R-waves are (or are conditioned to be) similar heights. It will probably do quite a bad job in a signal with a high T-wave 

DFT based filtering, than applying Gaussian derivative and Shannon energy  finally zero crossing detection on envelops, peak correction method  is best method for finding peaks for all type of signals including small peaks, signals corrupted with higher  amount of all noises.

I like to apply an overly sensitive peak detector, then go about discarding permutations of detected time-points  until RR variability over a window of time of interest is minimized. 

Try the iPhone/iPad App that I developed!

https://itunes.apple.com/app/ecg-wiz/id1317303333

https://www.youtube.com/watch?v=XVgjfMm5Lj8

The best and most widely used algorithm is Pan-Tompkin's. See the toolbox that I have recently developed, it implements several popular algorithms for QRS detection:

https://github.com/hooman650/BioSigKit

How can i convert edf file to .mat file