Detecting objects in images is a challenging task because of varying appearance and the wide range of poses of objects in images. There have been many methods implemented for this purpose but it still remains as an ongoing search area. In this assignment, we try to implement and evaluate histogram of oriented gradients (HOG) method which is one of the most successful recent object descriptors. It was created to allow the human form in images to be discriminated clearly at first then applied to other problem domains as well. We first study the issue of feature sets and parameters for object detection. We experiment HOG descriptor and linear SVM classifier using different parameters. We make several tests with different datasets. We try to detect different object types like bird, cat, cow, dog, horse, sheep, aeroplane that we extracted from pascal VOC 2007 dataset. We show that locally normalized Histogram of Oriented Gradient (HOG) descriptors sees the world from different kind of a perspective and when used wisely it gives perfect results for problem domains such as face detection and human detection.