Pedestrian Counting for a Large Scene Using a GigaPan Panorama and Exemplar-SVMs

A novel scheme for counting pedestrians in a large scene is presented in this paper. A panoramic imaging system, GigaPan, is employed to capture a high-resolution panorama that can cover a large area, such as a square, across 360 degrees. An improved object recognition method based on Exemplar-Support Vector Machines (SVMs) is used to detect pedestrians from the panoramic image. A histogram of oriented gradients (HOG) is used to characterize the objects. Due to the huge number of pixels in the high-resolution panorama, the time cost of the pedestrian counting is extremely high. A Graphics Processing Unit (GPU)-based scheme for parallel computation is adopted to reduce the time cost. The experimental results show that the proposed method is effective.