Real-time vehicle counting can effectively improve performance of traffic control and management systems. Counting vehicles in real time and with high performance under harsh conditions such as very crowded scenes or under different lighting is a challenging problem to be solved. With the use of deep learning approaches in the field of image processing, high-performance systems have been obtained. In this work, a novel 1-dimensional classification based deep learning approach is developed. The proposed system has high counting performance with very low computational load.