 This paper proposes an automated approach to detecting and counting cars in UAV images. First, the image is segmented into small homogenous regions, which are then used as candidates for car detection. A window is extracted around each region, and deep learning is used to extract features from the window. These features are fed into a CNN model, which is trained on auxiliary data, followed by a linear SVM classifier to determine whether or not the region contains a car. Finally, the region is filled in if it was determined to contain a car, and the process is repeated until all regions have been processed. Experiments were conducted on a challenging dataset of real UAV images, and the proposed method outperformed existing approaches in terms of accuracy and speed. This article was authored by Nassim Amour, Haqal Al-Hikri, Yaqoub Batzi, and others.