Motion Classification and Target Tracking

Target Tracking
DT（Temporal Difference）

Threshold T is about 15% of brightness, for 255 level, T = 40

We can calculate speed from Rn−1
Motion cropping.

Target Classification

Bi-variate classification:

In human shape, using morphological dilation. Segmentation method is Mahalanobis distance based.

Moving Target Classification and Tracking from Real-time Video

Feature Analysis
Motion feature extraction and analysis
direction and magnitude of motion.

1 and 8 as right;
2 and 3 as up;
4 and 5 as left;
6 and 7 as down.

motion magnitude calculation:

X and Y are directions correspondingly. We calculated average and biggest motion magnitude along X and Y direction for the whole video sequence.

Color and Edge Features
Median-cut algorithm to reduce the color map to about 256 colors, all pixels were back-mapped into homogeneous regions if the distance to the dominant color was not bigger than threshold to get regions corresponding to the first the 5 maximum dominant colors. Using a gradient edge operator to detect edges. Edge densities along different directions and different lengths around the dominant color region were calculated as edge features. Furthermore, we analyzed the edge information along four directions by distribution of visible edges and clustering them into horizontal, vertical and other category edge type.

Classification System

A new video clip is then classified as follows:following the tree, the feature which was utilized at Level 1 (the root level) is first extracted and the corresponding rule is applied, following which the path selected is chosen. At the next level, the same step is carried out whereby an appropriate feature is selected and the corresponding rule applied. In this system, only those relevant features are extracted and they are matched with the rule threshold directly. Further processing, such as data indexing, would be made right after the classification is done.

Rule-based Video Classification System for Basketball Video Indexing