Pedestrian detection plays a vital task in numerous computer vision applications, particularly in autonomous driving systems. These systems heavily depend on the perception module, which must be both high-performing and efficient in order to make accurate decisions in real-time with low delay. Prioritizing the prevention of collisions with pedestrians is vital in every autonomous driving system. As a result, pedestrian detection is a fundamental component of the perception modules in these systems. Recent years have seen a rapid development of the pedestrian detection system, which aims to alleviate difficulties caused by changes in illumination, scale, appearance, blur, and occlusion. However, existing techniques does not support all the challenges simultaneously with better accuracy in real-time. The proposed human detection algorithm addresses these challenges by utilizing scale generation architecture, which fuses features such as the histogram of oriented gradients (HOG) with non-maximum suppression (NMS), local binary patterns (LBP), local ternary patterns (LTP), and the gaussian mixture model (GMM). The scale generation architecture model supports the detection of humans at various scales. The HOG model, in conjunction with NMS, LBP, and LTP, extracts gradient edge features and fine-grained texture features, which support the pedestrian detection in a variety of appearances and illuminations. GMM improves feature performance by providing more useful information. We feed these features into the support vector machine (SVM) with a radial basis function (RBF) kernel to detect the pedestrian. We validate the proposed pedestrian detection system on more challenging state-of-the-art databases, namely INRIA and Caltech, achieving an average accuracy of 97.05% with real-time support that takes 0.12s to detect the pedestrian.