The recent emergence of deep learning algorithms has radically influenced the object detection performances by providing very precise and robust approaches. However, the disadvantage of the recently proposed object detectors is that they are computationaly extremely intensive, making them unsuitable for being used in portable devices. In this paper we propose six models which optimize an existent deep learning-based object detector - presenting also the results obtained on current benchmarks - for the particular task of person detection. We demonstrate that we may reduce the computational requirements of the existing algorithm by using several strategies for neural network architecture optimization, while still obtaining comparable results with the state-of-the-art approaches. Moreover, on the particular pedestrian detection task, we show that we matched state-of-the-art results.