The Drone is an unmanned aerial vehicle widely used to take pictures and record videos at high altitude, recording information for applications such as video surveillance, to be able to detect cars and people in real time, the main problem is that both the drone as objects are move, this make difficult the track objects with traditional techniques. Faced this problem, the present research proposes the use of convolutional neural network with multidomain learning (MDNet) and camera movement models for the detection and monitoring of cars based on aerial videos. The propouse obtaining very good results in compare with traditional methods, obtaining a 90 % of success in object tracking, which is useful for practical applications.