Attention Makers


Rotation Adaptive Visual Object Tracking with Motion Consistency

MAKERS: Litu , Deepak COUNTRY: India

The number of accidents can be reduced by estimating the trajectories of moving objects with the help of efficient and robust trackers based on the current position and velocity.


The Purpose

Visual Object tracking research has undergone significant improvement in the past few years. The emergence of tracking by detection approach in tracking paradigm has been quite successful in many ways. Recently, deep convolutional neural networks have been extensively used in most successful trackers. Yet, the standard approach has been based on correlation or feature selection with minimal consideration given to motion consistency. Thus, there is still a need to capture various physical constraints through motion consistency which will improve accuracy, robustness and more importantly rotation adaptiveness. Therefore, this project aims to solve these issues by incorporating various motion consistencies with rotation adaptiveness on top of tracking by detection approach.

The Technology

The core concept of this project is based on similarity measurement between image patches in the subsequent frames of a video with the sole supervision of the object in first frame.The similarity measurement task is carried by a CNN which is known as deep Siamese similarity learning network.The project includes a collaborative effort from deep learning and computer vision framework.The size(few KBs) and simplicity of the proposed architecture makes it compatible with smart phones.The integration of motion consistencies along with rotation invariance scheme have improved tracking performance in several challenging sequences,thereby enhanced the accuracy and robustness by a large margin.

Additional Details

=> Efficient tracking of aircrafts, missiles etc. to strengthen safety and security => Designing intelligent missiles with high robustness => Localization of particular objects such as land, water, vegetation etc. in satellite images using Deep similarity network => Driverless car => Surveillance camera => Automated traffic system => More effective segmentation using similarity learning => Illegal trespassing can be prevented with less supervision of human labour => Preventing accidents by estimating the trajectory of the moving objects such as pedestrian, vehicle, animal etc. => Detecting orientation of the object with high robustness will be a significant boost in the tracking community

Home Previous Next

Vote Share Comment