|Topic:||Convolutional Neural Networks for Loop Closure Detection in Visual SLAM Systems|
|Supervisor:||Luis Gomez Camara Garant: RNDr. Miroslav Kulich Ph.D.|
|Description:||We are looking for an extremely self-motivated student to work in the area of Deep Learning and Convolutional Neural Networks (CNN) with applications to Mobile Robotics. The ideal candidate will be exploring the capabilities of different CNN layers to extract meaningful information from images and to use this knowledge to identify features that are robust to image changes.
In particular, the student will be working on the problem of place recognition and loop closure detection under environment changing conditions (weather and season changes, day/night). His/her work will focus on the following main tasks:
- Transfer learning: Adaptation (fine tuning) of pre-trained off-the-shelf CNN's by feeding new samples representing the environment changes of interest
- Finding/collecting the image samples required for tuning the pre-trained networks
- Evaluation of the performance of the tuned networks in the problem of place recognition for loop closure detection
- Ability to prototype, test and iteratively improve computer vision algorithms
- Good knowledge of Linux, Python and OpenCV
- Understanding of CNN's and familiarity with frameworks such as Keras, Tensorflow, Pytorch, etc
- Excellent command of the English language, both written and spoken
This is a great opportunity to participate in prestigious research projects and the participant will get support to publish his/her results in recognized scientific journals/conferences. Based on performance, there is also potential towards PhD studies.