2. EVENT DETECTION
Event detection involves the automatic organization of a multimedia
collection C into groups of items, each (group) of which corresponds to
a distinct event.
3. CHALLENGES
1. requires application of several Computer
Vision
2. Involves subtleties that are readily
understood by humans, difficult to
encode for machine learning approaches
3. Can be complicated due to clutter in the
environment, lighting, camera placement,
traffic, etc.
6. EVENT DETECTION USING DATA MINING
TECHNIQUES
Video
Video Parsing and
Feature Detection
Instance Self Learning
Filtering and Reconstruction
Self Refining Training Dataset
Final Detection
Decision Tree Model
8. 3 BUILDING BLOCKS
1. Video Parsing and Feature Extraction
Involves temporal partitioning of the video sequence into meaningful units.
This module computes a large array of multimodal features (both visual and audio) from input videos
Five visual features are extracted for each shot:
1. pixel_change 2. histo_change;
3. background_mean 4. background_varr 5. dominant_color_ratio
2. Base Classifiers
Multiple base classifiers independently compute detection scores based on available features
3. Score Fusion
This module combines multiple base classifier scores through diverse fusion methods, and computes a single final
detection score per video clip
9. TWO- STEP PROCEDURE
1. Video content processing: The video clip is
segmented into certain analysis units and their
representative features are extracted.
2. Decision making: process that extracts the
semantic index from the feature descriptors.
10. DECISION MAKING PROCESS
DECISION MAKING
Knowledge Based Approaches
Rule based Classifier
Statistical Approaches
Support Vector Machines
Dynamic Bayesian Network
C4.5 decision trees
12. 2. VIDEO EVENT DETECTION BY INFERRING
TEMPORAL INSTANCE LABELS
Video recognition algorithm is inspired by
proportion SVM (p-SVM), which explicitly
models the latent unknown instance labels
together with the known label proportions
in a large-margin framework