A robust human detection system in an intelligent transportation system is desired by people and becomes essential to industries such as surveillance, automotive systems, and robotics. However, there are still many encounters to attain ideal human detection, such as the diversity of object appearance and the interference of an image due to light changing. These challenges make human detection a more challenging and unreliable task. Histograms of Oriented Gradients (HOG) are proven to be able to knowingly outpace existing feature sets for human detection. In this work, motivation only on the feature extraction method using HOG for real-time applications. For simplicity, a linear support vector machine (SVM) is used as a baseline classifier throughout the study. It is obvious that the calculation of HOG feature extraction is computationally complicated and unsuitable for hardware implementation. Hence, to adopt some approximate techniques, it will reduce implementation complexity and to improve extraction speed.