You can do this with MATLAB. Here is a nice example explaining it:

http://de.mathworks.com/help/vision/examples/detecting-cars-using-gaussian-mixture-models.html

This example is for a video but until Step 3 they explain it with a single image.

Phillip answer assume that you do have Matlab.

If you dont try with Open CV, its a great option and you will be creating not implementing:

http://www.intorobotics.com/opencv-tutorials-best-of//#opencv

Be creative, dont be lazy.

Hi Alireza,

If you work with Matlab, you can use this code:

http://mark-kay.net/2014/06/24/detecting-vehicles-cctv-image/

And for Opencv, you can use this code:

http://www.mathworks.com/help/images/examples/detecting-cars-in-a-video-of-traffic.html?refresh=true

These papers are so useful for your work:

http://www.jist.ir/UpFiles/OK/139402275-F.pdf

http://www.eecs.berkeley.edu/~malik/cs294/gavrila99.pdf

http://www.cs.bu.edu/~betke/papers/betke-mva00.pdf

http://www.pf.bgu.tum.de/isprs/cmrt05/pub/CMRT05_Gerhardinger_et_al.pdf

best,

If you have GPU then you can try YOLO Fresh out from the OVEN & and its real-time.

http://pjreddie.com/darknet/yolo/

Otherwise follow what Mario said

Ssd Model from Deep learning is highly recommended,

https://towardsdatascience.com/understanding-ssd-multibox-real-time-object-detection-in-deep-learning-495ef744fab

Idea is a bit abstract. A vehicle could be anything including two-wheeler. Thus, the shape varies. Complexity increases as vehicle is to be identified in a single image otherwise it was possible to track moving objects within comparatively static surroundings.

I would try with some edge detection schemes (Canny's etc.) to mark out regions followed by some initial image enhancement. I can then extract some shape/region features from various marked regions to investigate the presence of a vehicle. The system can first be trained on some prototype vehicles.