Do you already have an advisor at your school or university?
If so, you should talk with your advisor first.
It will be much easier for you to complete your research if you focus on areas your advisor already has expertise in. Look at what your advisor has published. What is their research interests? Could you align your own research interests with your advisors in some tangential way?
Not only will your advisor be more naturally interested in your research but they will be able to provide you additional help and guidance through their added domain knowledge.
If you don’t have an advisor I would suggest reading the most recent publications from popular computer vision and machine learning journals, including CVPR, ICML, NIPs, etc.
Read the abstracts and conclusions from these journals and keep track of the ones that interest you the most.
Then, when you’re done, go back to your short list and brainstorm how you might extend their works.