Weekly 2016.04.10

Week reported

1. Literature Reading

Mehran R, Oyama A, Shah m. Abnormal crowd behavior detection using social Force model[c]//Conference on computer Vision & Amp Pattern recognition. 2009:935-942.

This article introduces the use of social force models to detect abnormal behavior of people. At present, computer vision analysis of crowd behavior is a new field, such as computer automatic detection of crowd disturbances, chaotic behavior, and in the image to locate the location of the riots. At present, there are three main methods, the first is the microscopic method, defines the motive of pedestrian movement, the population behavior as a result of self-organization process, the social force model is the most famous representative of this method; the second is the macro method, which focuses on the goal-oriented crowd, he is not focusing on the individual, but divides the crowd into groups, Set the motion for each group; The third method is a hybrid model of the first two methods.

At present, the field of computer vision focuses on crowd behavior analysis. There are two ways to solve the difficulty of understanding crowd behavior, one is based on the object, think that the crowd is a separate individual composition, this method is to carry out the crowd segmentation or individual detection, but this method in the face of large density of people will have considerable difficulties. Another holistic approach is to treat people as a whole in large densities and to understand pedestrian behavior by capturing the behavior of groups.

Because of the crowding and density of the population, the object-based model evaluates the parameters of the social force model, so when evaluating the interaction force, the population is considered a combination of interacting particles, where the particles do not treat each individual as a particle, but rather the moving parts as particles, and the motion behavior of these particles constitutes the particle flow. Particle flow can capture the continuity of crowd flow, which is unmatched by other models.

Because of the instantaneous mechanical characteristics can not distinguish the abnormal phenomenon, through a period of time the mechanical model could be identified. For further processing of the image using LDA (latent Dirichlet Allocation), the Chinese name is the document theme generation model.

Through the image processing, the particle flow is further processed into the force field flow, the color can be used to identify the areas of greater strength, such as darker areas of color.

The author further compares the social force model with the optical flow model, and finds that the social force model has better effect than the optical flow model. The author concludes that the social force model is efficient and feasible in detecting and locating abnormal behavior.

Advantages: The author introduces in detail the mechanics model of some crowded crowd, including macroscopic, microscopic and mixed model, later image processing technology is I do not understand, read this article feel to do image processing is very interesting. At the same time, deepen the understanding of the social force model, the practical application of the social force model has some understanding.

Disadvantage: can only monitor the flow of people, identify and locate abnormal phenomena, can not be further based on the current situation of people to predict the flow of the next trend of people ahead.

2. Work progress

Help gong sister drying test pieces, measuring the weight of the specimen;

Complete part of the removal of particle module writing;

3. Next week's mission

Continue to read the literature;

Complete the program to remove the particle module and begin the initialization of the module;

Help the Gong sister to do the experiment.

Weekly 2016.04.10