This group on Video Analysis for Surveillance will be working on developing algorithms to analyze surveillance videos and images to draw inferences about a scene or situation. This will make use of video streams from one or multiple cameras to carry out analysis that helps in surveillance.
Personal Space ViolationAnalysis of human interaction in a social gathering is of high interest in security and surveillance applications. These interactions can be captured by analyzing the trajectories of the individuals over a time period. Here, in [1] we present a model to analyze all the trajectories together. This model can be used for predicting the future locations of the individuals. These predicted trajectories are then used to predict personal space violation for a candidate individual. The personal space for an individual is defined as a comfortable area around him/her and it gets violated if someone intrudes it. Here in this output, the objective is to identify if any intruder approaches the person in the center. When a violation is predicted, a warning is issued.
Group Activity AnalysisIn this work, our algorithm [2] predicts the activities at the individual, group, and collective levels. By an individual activity, we mean an action done by a person (walking or standing). To predict a group activity (activity performed by a group), we first discover the groups present in the crowd. The group activities are walking, crossing, waiting, queuing and talking. By collective activity, we mean a major activity happening in the scene. The categories are same as that of group activities.
Event Geo-LocalizationWe refer an event to be an object, for example, a concert or a parade which is video recorded by many smartphone cameras at the same time. Event Geo-localization refers to the task of computing 2-D location of the event on the surface of the earth. This is achieved with the help of the sensors, such as digital compass and GPS receivers available in modern smartphones. When multiple smartphone cameras video record the event, we capture this sensor data along with the video and use this auxiliary data (metadata) for event localization. Results on the collected experimental data shows that our approach in [3] is computationally very light and can be implemented in real time. Here, the objective is to localize the person on the surface of the earth as multiple cameras video record him.
Illumination Invariant TrackingTracking is a critical task in surveillance and activity analysis, where illumination changes can happen. The objective in this part of the project is to develop an illumination robust tracking framework. Towards this objective, we have developed two approaches. In [4] traditional mean-shift algorithm has been modified to be illumination invariant and in [5] a non-negative matrix factorization (NMF) based image model has been used to handle illumination changes. Results of the tracking algorithm in [5] is shown. Here, the green bounding box indicates that the person’s face is tracked even as the illumination changes.
Novel View SynthesisThe goal of Novel view synthesis is to synthesize new images from different viewpoints of given images. Here, Multiple views of an event and corresponding Meta-data of existing cameras are provided. In the results we generate how the event will look like from a given a new (imaginary) camera location.
Person Re-IdentificationPerson Re-Identification refers to matching pedestrian images across different camera viewpoints. Person Re-Identification becomes challenging due to non-overlapping camera views, changes in a person’s pose and occlusions. Here we use a Gaussian Of Gaussian (GOG) descriptor for pedestrian feature extraction and Cross-view Quadratic Discriminant Analysis (XQDA) based metric learning for Person Re-Identification. In the results, we show Top - 5 matches of a query pedestrian from two CCTV cameras.
|
|
Publications
[1] Neha Bhargava, Subhasis Chaudhuri and Guna Seetharaman, "Linear cyclic pursuit based prediction of personal space violation in surveillance videos", Proc. AIPR (IEEE Applied Imagery Pattern Recognition Workshop), Washington DC, Oct 2013.
[2] Neha Bhargava and Subhasis Chaudhuri, "Crowd motion analysis for group detection", Proc. ICVGIP (Indian Conference on Computer Vision, Graphics and Image Processing), Guwahati, Dec 2016. [3] Amit More and Subhasis Chaudhuri, "Event geo-localization and tracking from crowd-sourced video metadata", Proc. ICVGIP (Indian Conference on Computer Vision, Graphics and Image Processing), Guwahati, Dec 2016. [4] Kalyani Deopujari, Rajbabu Velmurugan and Kanchan Tiwari. "Spatial Temporal Weighted Histogram based Mean Shift Target tracking for Illumination Variation Format", Proc. ICVGIP (Indian Conference on Computer Vision, Graphics and Image Processing), Guwahati, Dec 2016. [5] Gargi Phadke, Shubham Dawande and Rajbabu Velmurugan, "Non-negative Matrix Factorization based Illumination Robust Mean-shift Tracking", Proc. NCC (National Conference on Communication), Chennai, March 2017. [6] Feroz Ali and Subhasis Chaudhuri, "Maximum Margin Metric Learning Over Discriminative Nullspace for Person Re-identification", Proc. ECCV (European Conference on Computer Vision) Munich, September 2018. |
People Involved in the Project
Faculty Members
Current Students