|
Mahesh ramachandran |
|
We present a computationally efficient and scalable technique for structure from motion using additional information about the motion of the camera. This additional information helps us to modify the SFM problem into a bilinear-like system and solve for the structure and motion in a decoupled fashion. We argue that the assumed additional information can be easily obtained using available sensors. We illustrate competent results on sequences collected from aerial and ground-based platforms. |
|
We present algorithms to perform stabilization and mosaicing of video sequences. We present algorithms for stabilization of low-resolution images. Then, we discuss how we can stabilize and mosaic sequences when we have additional metadata available along with a sequence. |
|
Project List |
|
We demonstrate results on moving object detection in aerial video sequences. We formulate the moving object detection problem as a joint tracking and classification problem. In this formulation, we are able to track a large number of features in the sequence robustly. This enables us to do better segmentation and reduce the number of false alarms in the sequence even for objects of small sizes. |
|
Networks of video cameras are deployed in many scenarios for surveillance and other uses. The huge amount of video data must be compressed as efficiently as possible for efficient storage. Traditional compression schemes exploit spatial and temporal redundancy for compression. We present a scheme for jointly compressing multiple sequences from different cameras with overlapping fields of view, by exploiting the inter-sensor redundancy. |
|
Distributed Video Coding |
