Human Detection in Still Images Using Hog with SVM: A Novel Approach

Abstract

Human detection in images gets more popular for applications includes night vision, robotics, surveillance etc. Existing systems in the literature support detecting single and multiple human objects. Algorithms developed for detecting humans in occluded condition have many troubles due to the problem of occlusion. This paper proposes a method to detect human objects in both multiple and occluded condition. To address the above problem, this method uses HOG for describing features and SVM for classifying objects. To examine the proposed algorithm, different set of images are used which consists of human and non-human objects like buildings, trees, lawns and roads. Then the features are obtained in the image dataset and assign the labeling to make a difference between human and non-human images. Thereafter feature vectors and labels are stored as a model for training the detector. SVM is used to train the detector. The feature vectors and labels are fed into the SVM to get trained. It cuts the image into slides and processes it for feature extraction. The detector uses the trained model for classifying new images for detection.