Direction Estimation for Pedestrian Monitoring System in Smart Cities: An HMM Based Approach

The paper proposes a novel approach for direction estimation of a moving pedestrian as perceived in a 2-D coordinate of field camera. The proposed direction estimation method is intended for pedestrian monitoring in traffic control systems. Apart from traffic control, direction of motion estimation is also very important in accident avoidance system for smart cars, assisted living systems, in occlusion prediction for seamless tracking in visual surveillance, and so on. The proposed video-based direction estimation method exploits the notion of perspective distortion as perceived in monocular vision of 2-D camera co-ordinate. The temporal pattern of change in dimension of pedestrian in a frame sequence is unique for each direction; hence, the dimensional change-based feature is used to estimate the direction of motion; eight discrete directions of motion are considered and the hidden Markov model is used for classification. The experiments are conducted over CASIA Dataset A, CASIA Dataset B, and over a self-acquired dataset: NITR Conscious Walk Dataset. The balanced accuracy of direction estimation for these experiments yields satisfactory results with accuracy indices as 94.58%, 90.87%, and 95.83%, respectively. The experiment also justifies with suitable test conditions about the characteristic features, such as robustness toward improper segmentation, partial occlusion, and changing orientation of head or body during walk of a pedestrian. The proposed method can be used as a standalone system or can be integrated with existing frame-based direction estimation methods for implementing a pedestrian monitoring system.