Why computers lose track of objects
Our brain is a genius at analyzing images that are communicated via the eyes. Without any effort, we can track a bird flying around in the woods or identify and follow our favorite football player during the game. However, this is far from obvious for a computer, as computer-vision experts know.
Object tracking by computers is extremely useful in applications such as video surveillance and human-computer interaction, robot navigation and autonomous driving. Today, videos in RGB or grayscale are used for object tracking. Problems arise when the scene is very complex: with a lot of background clutter, similar color of target and background, object rotation and deformation, ... Recent failures with autonomous cars are often related to these kinds of challenging scenarios for object tracking.
Hyperspectral video cameras can make object tracking more robust and reliable as they are able to identify an object much more thoroughly by using spectral fingerprints.
Hyperspectral camera used as object-tracking camera
At the 2021 edition of Whispers, participants are invited to join the ‘Hyperspectral Object Tracking Challenge’ organized by Griffith University (Australia) – one of the leading experts in the field of object tracking. In this competition, they are presented with a dataset of 40 training and 35 testing videos to try out their object-tracking algorithms. The contender with the best algorithms, wins the contest.
Imec, as an expert in hyperspectral video-rate cameras, was asked to optimize (the quality of) the training and testing videos. It was imec’s SNAPSHOT VIS camera that was selected to deliver datasets for the hyperspectral object-tracking challenge and whose images will challenge the participants in developing the best possible algorithms for object tracking.
Wouter Charle, Team Lead Software Development and Customer Interactions at imec Hyperspectral Imaging: “Our hyperspectral cameras are among the few that can deliver a high video rate. Moreover, they are very easy to use and – thanks to an improved data processing pipeline and newly developed algorithms – the quality of the spectral data is very good. For this reason, they can be used for a broad range of applications.”
Jun Zhou, Associate Professor at Griffith University: "The imec snapshot cameras are the only available hyperspectral product on the market that can fulfill a wide range of computer vision tasks, from object detection, recognition, classification to tracking."
Jun Zhou, Associate Professor at Griffith University: “We started to collaborate with imec in 2014 and have purchased several of imec’s hyperspectral cameras. The imec snapshot cameras are the only available hyperspectral product on the market that can fulfill a wide range of computer vision tasks, from object detection, recognition, classification to tracking. For the specific video tracking tasks, the imec snapshot cameras are the only product that offers high frame rate, high-quality hyperspectral images, and convenient operation. That's why we used the imec cameras to capture the dataset for the video contest.”
Let the (hyperspectral) games begin!
Want to know more?
13 October 2020