1. The Outline of the Proposed Method
This paper intends to present briefly about pedestrian attributes identification from the way of human walking. Walking is a daily activity, and it is one of the most pedestrian basic actions, and it can be observed regardless of public and private matters through all of space-time dimension. By seeing the human walking behavior, the various perceptions of body conditions can be easily observed when abnormality is found in the limbs, the human motor coordination, and the human sensory function. The pedestrian posture and walking speed condition in which the defective function is found, are reflected from those scenes.
In this research, space-time pattern information is extracted based on the dynamic scene acquired from a video image at the pedestrian’s crossing area. The walking condition of person which might represent the perception of body modality differences between “normal-walking person” and “abnormal-walking person” is evaluated from the dynamic features of walking pattern. Then, an intelligent information processing mechanism that embedding the aspect of the human thinking way connected with the factor of the walking appearance is determined by the soft computing approach.
2. Understanding Pedestrian Walking Behavior.
The construction of an autonomous supports system is required to be developed together with welfare and the medical treatment facilities. The purpose is to keep safe our society in their daily activities, coincide with the falling birthrate and aging problems, especially, in Japan. It is important to introduce the surveillance system in private/public spaces in which physically/mentally dependent, e.g., the senior citizen, the children, and others challenged person are involved inside of that environment. The system should be proficient to detect abnormality behavior of person autonomously, and simultaneously manage the content of abnormality, the grasp, and the decision under emergency situation for preventing incident.
The walking behavior of person can be as a product generating by human motorical systems. It is possible that the walking behavior condition in certain circumstances becomes unexpected to the people who have deteriorated physical or physiological condition such as the aged or others challenged person. Based on this state condition, the feature different of “Walking Mode” between those groups and a usual adult might be observable and predictable. Thus, the determination of pedestrian attributes can be developed through understanding pattern of human walking behavior.
Automated determination of pedestrian attributes from motion scenes by the intelligent systems, in general, require two processes, i.e., extracting the relevant features from video and machine learning capability before the recognition are to be performed. The first process is required to extract the features which representing the human attributes. The second process is required to build a system which is able to recognize similar behaviors based on given learning patterns. It can be attained by macroscopic information fusion mechanism based on qualitative features of space-time motion pattern of walking pedestrian.
3. Technology Requirement.
Determination of pedestrian attributes based on their behavior might be difficult to be obtained by the conventional approaches such as neural networks. The aggregation mechanisms of each neuron in neural networks are simple weighted sum which might not be used effectively for the problems where complex determining processes are essentially required. In addition, the extracted features, in most cases, are qualitative and they may be captured by fuzzy categories. Thus, the method which is able to emulate human processing thinking-way is required to be developed.
In our study, Choquet Integral Agent Networks (CHIAN) was introduced as macroscopic information fusion mechanisms because of their flexible integration of multiple qualitative input data. In addition, CHIAN has information fusion mechanism as human mimetic thinking-way in which each agent has to assign the corresponding connection strength or weight to the subset of input units based on the presented pattern. Each agent of CHIAN has also a meaning which might be similar to embedding human tacit knowledge. Based on that fact, CHIAN might be effectively employed for solving the problems in which human mimetic thinking ways are involved in the process of extraction and information fusion mechanism.
Detecting human attributes from pattern of human motion requires, first, representation feature which is able to discriminate a class or a group based on the presented pattern; however selecting and extracting an appropriate feature is still part of art rather than science. In this stage, deep impression of human motion mechanism should be included in initial processing such as feature extraction process. Second, aggregation information mechanism is required to solve the given problem. However, current machine learning technology is still behind of human capability when make a decision. From this point of view, development of machine learning method which is able to imitate human thinking-way is necessary to be developed, especially for coping with the problem where complex determining processes are required.
References
- J.J. Little, J.E. Boyd, “Recognizing People by Their gait: The Shape of Motion,” Journal of Computer Vision Research, The MIT press, Vol. 1(2), 1998.
- 2. J.K. Aggarwal and Q. Cai,” Human Motion Analysis: A Review,” Computer Vision and Image Understanding, Vol. 73(3), pp. 428-440, 1999.
- Mubarak Shah, “Understanding human behavior from motion imagery,” Journal on Machine Vision and Applications, Vol. 14(4). pp. 210-214, September 2003.
- Handri Santoso, “Determination of Pedestrian Attributes from Motion-Imagery Monitoring Using a Soft Computing Approach,” Doctor Dissertation, Nagaoka University of Technology, August 2008.