Type: simulation
Download 14.44 Kb.
|
| Paper: Crowd-Z: the user-friendly framework for crowd simulation on an architectural floor plan
Type: simulation Summary: Paper propose a tool for interactively design simulation scenarios by drawing floor plans and preferred routes. Paper: Performance Evaluation of Crowd Image Analysis using the PETS2009 Dataset Type: Survey/Review dataset Summary The paper revise the history of performance evaluation and performance measures through the PETS dataset, a well known dataset for computer vision tracking problem. The dataset includes mid-level crowd videos and high density only for few clips. It introduces the problem of evaluating crowd detection and tracking accuracy and revise many methods in literature. I propose this as the first paper as a survey paper. Paper: An agent-based crowd behaviour model for real time crowd behaviour simulation Type: Simulation Summary: The paper proposes an agent based simulation approach that uses both local social forces and a global hydrodynamics approach. The approach is coupled with intensive large scale processing and an algorithm to perform real time simulation when a mid number of agents is present is depicted. Results are very satisfying. Paper: Mobility Analysis of the Aged Pedestrians by Experiment and Simulation Type: Simulation / Experiments on real data Summary: The paper is about experiments for characterizing the motion of aged pedestrian in crowd. The objective is to grasp important feature that can be combined in a cellular automata model to effectively model pedestrian motion. Paper is experimental and very good. Paper: Activity Analysis in Crowded Environments Using Social Cues for Group Discovery and Human Interaction Modeling Type: Pattern Recognition/Vision Summary: The paper is about discovering interacting group in videos and classifying the type of interaction. This is performed by creating a graph where pedestrian are graph nodes and clustering graphs in order to find the groups. Every group is assigned to a feature vector composed by group members motion and relatives position. This feature are used for group activity classification using a discriminative classifier and bag of words. Paper is straightforward but effective Paper: Scene Invariant Multi Camera Crowd Counting Type: Pattern Recognition/vision Summary: The Paper is about crowd counting in a multicamera environment. It performs camera calibration and use multiple views to overlap the projections of the pedestrian. Test are on the PETS09 dataset. Not outstanding quality but experiments are robust and complete Paper: Real-Time Video Event Detection in Crowded Scenes using MPEG Derived Features: a Multiple Instance Learning Approach Type Pattern Recognition/vision Summary: The paper is about a real time approach for event detection in crowd scenes. It uses Mpeg motion vectors and trajectories to describe the motion amount of the scene. Finally the classification is based on a sparse Multiple instance learning approach that achieves real time performance. Very interesting paper with broad applicability. Paper: Unsupervised Dense Crowd Detection by Multiscale Texture Analysis Type: Pattern Recognition/Vision Summary: The paper propose a method for detection of dense crowd by texture analysis. A Multi scale analysis of textural features is performed using both color (HSV and RGB) and LoG (Laplacian of Gaussian for textures.). Eventually a graph with multi scale feature is computed and clustered into two cluster crowd/non-crowd. Finally segmentation is applied. Naïve method very effective. TEST on BOTH SIMULATION VIDEOS AND IMAGES. Paper: Towards an Integrated Approach to Crowd Analysis and Crowd Synthesis: a Case Study and First Results Type: Simulation Summary Experiments and test about crowd formation and group shapes. Very Interesting and of general comprehension. First paper of simulation part in my opinion. Paper: Bayesian Human Motion Intentionality Prediction in Urban Environments Type: Pattern Recognition/Vision Summary:
Paper is about a motion prediction method for pedestrian in crowd using a Bayesian graphical model in order to compute the most probable motion vector depending on several factors: surrounding people, entry and exist zones, previous direction…. All is mixed by enforcing Gaussianity in order to make the model tractable. Experiments are performed on Video Datasets and with a mobile robot that is guided by the Bayesian predictor to crowd. Average paper. Paper: Contextual Anomaly Detection in Crowded Surveillance Scenes Type Pattern –recognition/ Vision Summary:
The paper is about detecting anomalies in crowd videos using trajectory information. First a state of the art tracker is used for extract trajectories. Then mutual information among motion trajcetories is used as the main feature for classification. Authors define an ontology of anomalies depending on the context and introduce the context in the classification model. Test are performed on PETS dataset. A very good paper. Paper: A bio-inspired multi-camera system for dynamic crowd analysis. Type: simulation Summary:
Paper is a bit beyond my expertise. They propose to seolve the art gallery problem where guard are replaced by moving cameras. Roblem here is to place nd move cameras in order to obtain the maximum coverage of the observed area. This is achieve by a spider approach borrow from bio-informatics. The paper is well written and very detailed. Paper: Improving Counterflow Detection in Dense Crowds with Scene Features Type: Pattern Recognition/Vision Summary:
Paper is about counterflow moving detection by visual features analysis. A scene heat map is computed to produce probable counterflow zones. Then a racking algorithm is employed to finely track the people that moves counterflow. Effective paper but not outstanding. Paper: Using the AGORASET dataset : assessing for the quality of crowd video analysis methods Type: simulation dataset Summary:
Paper proposes the AGORASET a computer graphics dataset for simulating pedestrian and obtain videos with ground truths for performance analysis. Very nice and interesting. Paper: Abnormal Behavior Detection Using Hybrid Agents in Crowded Scenes Type: simulation/Vision Summary:
The paper is in between to simulation and vision. It observe a video and divide the video into overlapping blocks. Then an agent is assigned to every block. Agents evolve on the basis of the social force model and optical flows. Agents responses are then combined in a feature vector to classify the behavior either in normal or abnormal. Nice paper but very simple. Paper: Motion Feature Filtering for Event Detection in Crowded Scenes Type: Pattern Recognition/Vision Summary:
The paper is about scene classification and event detection in videos. First motion features are computed by optical flow and then filtered in order to obtain a smooth response. Features are arranged in a motion features vector that identify the scene and classified according to a probabilistic classifier. Here the key of the work is the motion smoothing procedure that exhibits very interesting robustness against conventional motion vectors. Directory: files2 files2 -> Осетрова Е. Е. Пособие по общественно-политическому переводу elections files2 -> Minutes of the emc save the Rain Meeting Download 14.44 Kb. Share with your friends:
|