Human detection using speeded-up robust features and support vector machine from aerial images

Human detection from an aerial image has attracted wide attention due to its vast area of application such as in surveillance, search and rescue operation, and for visual understanding of the image. Unlike object detection, human detection from an aerial image is a challenging classification problem because of different posture appearance of human in an image. More so, at high altitude human shape appear deformed. Different features selection and different algorithm have been proposed. Although effective, but limited due to, characteristic of human posture in an image. In order to address this problem, this research proposed a Speeded-Up Robust feature selection and SVM for human detection from an aerial image due to computational speed and robustness of the SURF feature. This approach would help in better human detection from aerial images irrespective of position and movement for either rescue or surveillance mission. Aerial images were acquired preprocess and segmented using Otsu segmentation. A database comprises of two hundred images was created; 70 percent (140 images) of it was used in training the classifier and 30 percent (60 images) for testing the classifier. Accuracy of 50%, specificity of 57.1%, sensitivity of 46.2% and precision of 66.7% was achieved. These results can be used for a better human detection from an aerial image irrespective of the position or movement.