There is no way to do it. A neural network does not actually detect an object. If just classifies vectors describing images into classes. However, the object itself is never recognized. The only thing that is recognized is the similarity or dissimilarity between two images. To achieve what you want, you need to use an object recognition method. Take a look here:
http://www.intorobotics.com/how-to-detect-and-track-object-with-opencv/