Mahesh ramachandran

Motion detection

We detect independently moving objects from aerial videos using telemetry information in the form of metadata available along with the sequence. We propose a joint feature tracking and segmentation strategy to solve this problem. More information on this will be disclosed at a later date.

 

Please find an illustrative video sequence below. This sequence illustrates a mosaic of a long sequence from the VIVID dataset (around 1200 frames). A joint feature tracking strategy is used to track background features and compute the inter-image transformations. The moving objects are segmented and removed from the sequence.  There are some artifacts due to errors in segmentation (work on improved segmentation is in progress).