My robot needs to obtain visual odometry, with a condition that the used algorithm has to have as low computational requirements as possible and the used camera/sensor shouldn't be very expensive. The thing is that it'll run on a platform like Jetson together with lots of other CPU expensive processes.
Currently I'm trying monocular odometry packages from ROS (fovis, svo...) with an ASUS Xtion sensor which doesn't give very good results (due to computational power with fovis or probably insufficient FOV with svo).
As I don't have a stereo camera to compare it with and I don't want to buy one until I don't know if it helps, I want to ask whether stereo algorithms would be a better choice here (are they more effective?), having in mind the low-cost budget and low cpu requirements. The robot also has to keep the depth sensor (doesn't have to be strictly ASUS Xtion), so the possibilities are to use a different sensor usable for obtaining visual odometry or to equip the robot with another camera. If you have any other suggestions I will appreciate it. Thank you in advance!