• ●  Built both location classifier and scale classifier.
• ●  Built parallel pipeline for multiple targets tracking.
• ●  Conducted test on overlap targets tracking.
• ●  Combined Appearance-based tracking with Kalman filter.
• ●  Added a work-around approach for handling occlusions.
• ●  Calculated and predicted the motion and position when occlusion is assumed.

Parsing Point Cloud into Disjoint Spaces

● Detection incorporating void spaces
○ Detecting the peak-gap-peak pattern ○ Merging

● Canonical Coordinate System Among Spaces
Semantic blocks to pixel level boundary segmentation

• ●  Enforcing Contextual Consistency using CRF
• ●  Updating the Disjoint Space Parsing Results