@ArrowLLL
2017-12-08T13:23:00.000000Z
字数 9474
阅读 1503
OPTIMAL
行人行为分析与人群行为分析是视觉场景理解的两个重要分支，但在论文当中基本是混在一起的，并不加以区分。其主要应用于带有时间性质的视频应用当中。
就一个星期的探究，主要对这个方面的应用做出一点总结。
总共有四个大的方面，一是智能环境设计与规划，二是视觉应用设计与仿真，三是密集环境个体导航，四是视频监控及检测追踪。
智能环境设计方面，Forecasting Interactive Dynamics of Pedestrians With Fictitious Play（CVPR 2017）中描述的交互式自治系统(interactive autonomous systems)，用于理解不同的个体与人群行为并且对预测到的动作与行为做出响应，如自动驾驶，家庭机器人，智能家居等。聚焦于人类和自治系统（机器人等）在共享物理环境中的交互。
Crowd analysis: a survey （from 《Machine Vision and Applications》2008）当中提到人群分析也可以为公共空间的设计提供指导。例如，设计购物商场的布局使其对顾客更加方便，或者最大化一个办公室的空间设计等等。
在Crowds by Example （from 《Computer Graphics forum》2007） 当中提到，计算机仿真生成的人群在电影、电脑游戏和其他虚拟世界应用于仿真中变得越来越普遍。随着这些方面的应用对高质量现实感和场景复杂性越来越严苛的要求，真实可信的人群仿真需求也变得越来越大。尽管有灵气的单人动作的模拟方法已经取得极高的成就，但创造一个真实可信的人群仍然是一个巨大的挑战。
密集环境个体导航应用主要分为两方面，一个是应用于个体的小型交通工具的导航，如平衡车，电动购物车等，一个是机器人在密集环境中的个体导航。
在Learning Social Etiquette: Human Trajectory Understanding In Crowded Scenes （ECCV 2016） 和 Forecasting Social Navigation in Crowded Complex Scenes （from arxiv.org） 中提到，当行人和骑自行车的人在密集环境如校园、购物商场等空间穿行时，都会基于一些"社会"规则（如靠右行，避开车流等）互相配合尽量减小干扰。这些可以通过建立模型学得这些规则与交互，应用于个体导航。
在Planning-based prediction for pedestrians （from IROS 2009. IEEE）中说到，在人流当中确定机器人的运动已经well-studied但仍然很困难。机器人避免碰撞的方法总是很难让人满意，机器人应当学会预测行人的位置然后做出相应的计划避免碰撞。
由于逐步提高的安全需求，视觉视频异常检测(video abnomaly detection)在智能视觉监控领域已经变成了一个很重要的研究领域。在Video anomaly detection based on a hierarchical activity discovery within spatio-temporal contexts （from 《Neurocomputing》, volume 143, 2 November 2014, Pages 144-152） 中说到，传统监控方式的弊端逐渐显现，如工作人员长时间凝视屏幕引起的疲劳和注意力不集中，使得他们忽略了潜在的危险事件，另外，已经存在的监控主要用于调查取证，并不能在危险事件发生时发出警告。已经有许多研究人员对这些方面进行研究并提出了自己解决相应问题的方法，如基于对隐藏视觉属性建模的无监督方法和基于训练模型或者已标记行为模板的监督方法。无论哪一种，异常行为描述的界定和场景复杂度都使得异常行为检测成为一个具有挑战性的问题。
from "CVPR 2016"
socially-aware robots
For many tasks in populated environments, robots need to keep track of current and future motion states of people.
people tracking is a key technology for mobile robots to be safely and efficiently deployed in populated environments.
design of intelligent tracking systems
Crowd behavior analysis : automatic detection of riots or chaotic acts in crowds and localization of the abnormal regions in scenes for high resolution analysis
from《Neurocomputing》
target tracking[4]
Detecting and tracking people in crowded scenes is a crucial component for a wide range of applications including surveillance, group behavior modeling and crowd disaster prevention.
video anomaly detection5
Video anomaly detection has become an important research aspect in the area of intelligent visual surveillance due to growing security needs. With the application of video surveillance in modern life growing gradually, drawbacks of conventional surveillance are revealing themselves; for example, spending a long time staring at monitors causes operators fatigue and inattention, sometimes causing them to neglect certain underlying dangerous occurrences. Additionally, since existing surveillance functions tend to capture evidence in a surveying manner, they cannot provide a warning when risk events are forming. Although automated anomaly-detection has attractive potential, it is also one of difficult problems in video analysis. Firstly, unusual events are rare, difficult to describe, and often subtle; secondly, visual behavior is diverse and complex in a realistic and unconstrained environment; thirdly, the description and definition of normality and abnormality, have high uncertainty and depend on changing visual contexts.
Counting crowd flow is a video-frame analyzing process, which uses computer vision techniques to estimate the crowd information. It has been widely applied in fields such as public security, urban public transport, resource allocation, and optimization.
Humans navigate crowded spaces (introduction)
When pedestrians or bicyclists navigate their way through crowded spaces such as a university campus, a shopping mall or the sidewalks of a busy street, they follow common sense conventions based on social etiquette.
crowd simulation[1-5]
Computer generated crowds are thus becoming common in films, computer games and other virtual world applications and simulations. As these applications continue to strive towards higher levels of realism and scene complexity, there is an increasing need for realistic and believable crowd simulations.
pedestrian behavior modeling
- Architects are interested in understanding how individuals move into buildings to create optimal space designs.
- Transport engineers face the problem of integration of transportation facilities, with particular emphasis on safety issues for pedestrians.
- Recent tragic events have increased the interest for automatic video surveillance systems, able to monitor pedestrian flows in public spaces, throwing alarms when abnormal behavior occurs.
- Special emphasis has been given to more specific evacuation scenarios, for obvious reasons.
determining robot movements
Robots should certainly never collide with people 11, but avoiding collisions alone is often unsatisfactory because the disruption of almost colliding can be burdensome to people and sub-optimal for robots. Instead, robots should predict the future locations of people and plan routes that will avoid such hindrances (i.e., situations where the person's natural behavior is disrupted due to a robot's proximity) while still efficiently achieving the robot's objectives.
Human re-identification is a fundamental and crucial problem for multi-camera surveillance systems. It involves re-identifying individuals after they leave field-of-view (FOV) of one camera and appear in FOV of another camera.
automatic re-identification in dense crowds will allow successful monitoring and analysis of crowded events.
When pedestrians walk in a crowded space such as a university campus, a shopping mall or the sidewalks of a busy street, they follow common sense conventions based on social etiquette
We believe the task of visual prediction is important for two main reasons
- (a) For intelligent agents and systems, prediction is vital for decision making. For example, in order to perform assistive activities, robots must be able to predict the intentions of other agents in the scene. Even a task as simple as walking through a crowded hallway requires the prediction of human trajectories.
- (b) More importantly, prediction requires deep understanding of the visual world and complex interplay between different elements of the scene. Therefore, prediction can act as a way to define “what does it mean to understand an image,” and the task of visual prediction can act as the litmus test for scene understanding.
We focus on predictive models since they are important for developing interactive autonomous systems (e.g., autonomous cars, home robots, smart homes) that can understand different human behavior and pre-emptively respond to future human actions.
profiling crowd attributes
crowd density, collectiveness and stability
behavior understanding
abnormal event detection, crowd scene classification and crowd behavior recognition
Crowd management
Crowd analysis can be used for developing crowd management strategies, especially for increasingly more frquent and popular events such as sport matches, large concerts, public demonstrations and so on, to avoid crowd related disasters and insure public safety
Public space design
Crowd analysis can provide guidelines for the design of public spaces, e.g. to make the layout of shopping malls more convenient to costumers or to optimize the space usage of an office.
Virtual environments
Mathematical models of crowds can be employed in virtual environments to enhance the simulation of crowd phenomena, to enrich the human life experience.
Visual surveillance
Crowd analysis can be used for automatic detection of anomalies and alarms. Furthermore, the ability to track individuals in a crowd could help the police to catch suspects.
Intelligent environments
In some intelligent environments which involve large groups of people, crowd analysis is a prerequisite for assisting the crowd or an individual in the crowd. For example, in a museum deciding how to divert the crowd based on to the patterns of crowd.
Pedestrian travel time between entrances and exits indicates traffic efficiency and travel cost of a scene, and thus attracts great attention in surveillance applications. When the travel time increases due to scene congestion, security administrators can take prompt actions, such as blocking some entrances until the congested crowds disperse, or opening extra exists, to control traffic.
Travelers can also use such information to make plans.
Travel time itself can be also regarded as an important feature to describe each individual's behavior and determine whether a pedestrian behaves normally or not.
Multiple object tracking (MOT) has a long tradition for
applications such as radar tracking
Crowd management
Crowd modeling and analysis can help comprehending, thus managing, public traffic and gatherings, as well as related events.
Urban planning
Understanding the undergoing interactions in the crowd as well as the behaviors of individuals in common places can assist in designing the structural layout of public spaces in order to accommodate the different crowd mobility flows.
Security and risk management
Monitoring public masses for the aim of security and hazard prevention is one of the top priorities with regards to today’s society. The automatization of such process is therefore pivotal to aid ensuring secure and smooth daily activities. Moreover, detecting abnormalities may even help alarming yet preventing potential future threats. In turn, time, cost and human labor can potentially be saved.
recognition of potential risks in video surveillance, etc.