Head Pose Classification in Crowded Scenes

Human head pose and gaze directions have traditionally been studied for expression and face recognition, and human computer interaction [6]. Accurate estimations of either head or gaze direction can provide useful information for the inference of a person’s intent and behaviour. However, most existing techniques rely upon medium to high resolution images captured under well controlled conditions from a fairly close distance [3, 5, 7, 10]. Given high resolution images, most existing techniques deploy extensive feature extraction to capture detailed head/facial shape and texture information. However, this approach relies on accurate subtraction of head foreground region from the background which is not always feasible. We propose a novel technique for head pose classification in crowded public space under poor lighting and in low-resolution video images. Unlike previous approaches, we avoid the need for explicit segmentation of skin and hair regions from a head image and implicitly encode spatial information using a grid map for more robustness given low-resolution images. Specifically, a new head pose descriptor is formulated using similarity distance maps by indexing each pixel of a head image to the mean appearance templates of head images at different poses. These distance feature maps are then used to train a multi-class Support Vector Machine for pose classification. Our approach is evaluated against established techniques [2, 8, 9] using the i-LIDS underground scene dataset under challenging lighting and viewing conditions. As in [8], a 360◦ head pose in panning angle is discretized into eight pose classes with 45◦ increment. The results demonstrate that our model gives significant improvement in head pose estimation accuracy, with over 80% pose recognition rate against 32% from the best of existing models.