Communications of the ACM
Scientists Adopt Deep Learning for Multi-Object Tracking
Deep-learning-based methods make the handling of high-dimensional appearance information feasible.
Researchers at the Gwangju Institute of Science and Technology (GIST) have adapted deep learning techniques in a multi-object tracking framework, overcoming short-term occlusion and achieving remarkable performance without sacrificing computational speed.
Despite steady progress in computer vision over the past decade, some tasks are still difficult for computers to perform with acceptable accuracy and speed. One example is object tracking. While computers can simultaneously track more objects than humans, they usually fail to discriminate the appearance of different objects. This can lead algorithms to mix up objects in a scene and ultimately produce incorrect tracking results.
At GIST, a team of researchers addressed these issues by incorporating deep learning techniques into a multi-object tracking framework. They present a new tracking model they call 'deep temporal appearance matching association' in "Online Multiple Pedestrians Tracking Using Deep Temporal Appearance Matching Association," published in Information Sciences.
The tracking model enables a comparison of two inputs to be used for adaptive appearance modeling and contributes to the disambiguation of target-observation matching and to the consolidation of identity consistency.
From Gwangju Institute of Science and Technology
View Full Article
No entries found