Real Time Abnormal Motion Detection in Surveillance Video Background Motivation Examples Constraints & Requirements

Real Time Abnormal Motion Detection in Surveillance Video Background Motivation Examples Constraints & Requirements

Real Time Abnormal Motion Detection in Surveillance Video Background Motivation Examples Constraints & Requirements

Fitz, Dave, News Director;Morning Show Host has reference to this Academic Journal, PHwiki organized this Journal Real Time Abnormal Motion Detection in Surveillance Video Nahum Kiryati Tammy Riklin Raviv Yan Ivanchenko Shay Rochel Vision in addition to Image Analysis Laboratory School of Electrical Engineering Tel Aviv University Background Huge video surveillance system can apply hundreds in addition to even thous in addition to s of cameras. To avoid communication bottlenecks, the acquired video is often compressed by a local processor within the camera, or at a nearby video-server. The compressed video is then transmitted to a central facility as long as storage in addition to display. Motivation Difficulties Extensive human monitoring of the incoming video channels is impractical, expensive in addition to ineffective Storing a large capacity of data is problematic Solution Automatic systems that trigger recording or video transmission in addition to attract the attention of a human observer to a particular video channel. Static scenes – Motion detection Dynamic scenes – Abnormal motion detection

Universidad Panamericana GT

This Particular University is Related to this Particular Journal

Examples Airports L in addition to transportation terminals Roads Office buildings Private usage Constraints & Requirements Real time operation High reliability Low cost hardware Limited computation power Related work Motion Segmentation in addition to abnormal motion behavior Ermis et al ICIP 2008

Key Ideas Avoid segmentation or tracking Use the macro-block motion vectors that are generated anyway as part of st in addition to ard video compression methods Derive motion features from the motion vectors Estimate the statistical distributions of the motion feature vectors that characterize normal activity during training Unlikely feature vectors during online operation indicate abnormal motion. From Video to Motion Vectors Video compression – Eliminating spatial & temporal redundancy Intra-frame – compressed as a full still image Inter-frame – represented by macro-block displacement vectors (motion vectors) relative to the reference frame in addition to an error image. From Motion Vectors to Motion Features Construct a small set of features. Define in addition to as the magnitude in addition to the direction of a motion vector

From Motion Vectors to Motion Features Total absolute motion Divide the frame into sub-regions 2. Area of dominant motion 3. Motion homogeneity From Motion Vectors to Motion Features The range of motion directions is divided into equal fractions of size . Let be the angular fraction index 4. Principal motion direction 5. Dominance of principal motion direction Manually designed features Machine learning algorithms (such as Boosting or SVM) learn set of discriminative features (or support vectors) during training. The `learned’ set of features is adapted to the scene. Why Negative examples (in addition to the positive ones) are a must! It is impossible to train the system on a comprehensive set of abnormal motion patterns as long as every possible scene. Why not

Probability Density Estimation Construct an dimensional histogram of the discrete feature vectors obtained during the training. The normalized histogram defines the PDF of normal motion under the assumptions: The training detects only normal motion The training captures the variability of the possible normal motion patterns. Histogram Why Why not Parametric methods as long as density estimation allow compact representation using statistical measures such as mean, variance in addition to likelihood. The data remains continuous. Quantization (resolution) is not an issue. There is no reason to assume parametric distribution of the data Histograms are the fastest in addition to simplest non-parametric estimation methods. Abnormal Motion Detection Set a threshold to define the unlikely feature vectors (histogram cells). Compute the feature vector associated with each of the incoming frames at the operational phase. The system detects abnormal motion event after consecutive unlikely feature vectors.

Abnormal Motion Detection Event graph Experiments Snapshot of the operational system

Experiments Some technical details 50 minutes of video were acquired 41 minutes of normal traffic were used as long as training The 9 minutes test sequence contained normal in addition to abnormal motion The movie was captured using SONY TRV900E PAL (25 pfs) digital video camera It was trans as long as med to DV as long as mat an coded to MPEG-1 using generic MPEG-2 codec Summary The system presented is: Computationally efficient (75 CIF frames/sec on 2.8GHz PC) Reliable Operates on the compressed video stream Abnormal motion is not associated with a particular object in the scene Possible extensions – relate the concepts of normal in addition to abnormal motion with time in addition to causality Questions The complete 9 minutes test movie can be found at: We thank I. Dvir in addition to D. Harari as long as the stimulating discussions.

Fitz, Dave WKXX-FM News Director;Morning Show Host

Fitz, Dave News Director;Morning Show Host

Fitz, Dave is from United States and they belong to WKXX-FM and they are from  Rainbow City, United States got related to this Particular Journal. and Fitz, Dave deal with the subjects like Entertainment; International News; Local News; Music; National News; Regional News

Journal Ratings by Universidad Panamericana

This Particular Journal got reviewed and rated by Universidad Panamericana and short form of this particular Institution is GT and gave this Journal an Excellent Rating.