Real time object tracking commonly suffers from various real life problems like varying pose, occlusion and illumination.In this paper, the illumination problem is handled that highly influences the quality of people detection.The main focus is to eliminate the lighting effect and increase the success ratio in detection.The proposed system removes the noise from single-scale retinex (SSR) image via linear smoothing filter which in fact segregates the actual features from illumination information.In addition, we extend processed images with original histogram of oriented gradients (HOG) based detector and local binary patterns (LBP) for training to classify the negative and positive labels.Moreover, we utilized support vector machine (SVM) for classification purpose.Comparison results show that the proposed method have more accuracy as compared to original HOG+SVM method.