Lidar measures distances and angular vectors to each point directly, so in comparison, and from there it is relatively easy to filter the points ( like by first return ) and sometimes the return pulse can even be signal processed to distinguish other attributes about the surface. A lot of this processing can be pipe lined and done more or less in parallel ( each calculation can be done interdependently of the others ) until the very end, and even the surface generation is amicable to great efficiency. SFM on the other hand ( over simplification ) is orders of magnitude more involved, the relationships between the pixels in and image need to be compared to one another, and then compared to every other image, and ambiguities resolved ( https://towardsdatascience.com/structure-from-motion-311c0cb50e8d ). The speed this can occur is greatly affected by the image complexity and instrument setup ( single static camera multiple shots ) moving single camera, multiple positioned fixed cameras, etc. ). Basically, it can happen as fast as the amount of money available to throw at the problem, i.e. if you're Hollywood shooting a SFX scene on a set with high rez cameras connected to super high bandwidth with massive connection to a VM with unlimited RAM on a computer cluster in the cloud, it's totally doable in real time.

Hi ,

The privilege of Lidar in this field is precision & higher frequency (rate) due to less information, but their are limited to odometry not modeling the environment as usually desired in SFM or Photogrammetry software.

I guess you haven't familiar with SLAM ! (Simultaneous localization and mapping)

The Real-time visual SLAM algorithm ( in some cases visual odometry) is exactly what you're looking for.

Common algorithm made a feature base map of environment but few method have good mapping outputs. ( cause most of them created for Odometry)

you can Search by SLAM , Visual Odometry in YouTube again to see the the results.

there are some public implementation that you can easily use . like SVO which is implemented in ROS (Robotic operating system).

also there are some other projects with different outputs and etc. (like BundleFusion)

https://www.youtube.com/watch?v=keIirXrRb1k

https://graphics.stanford.edu/projects/bundlefusion/

https://github.com/raulmur/ORB_SLAM2

http://youtu.be/2YnIMfw6bJY

http://rpg.ifi.uzh.ch/svo2.html

all the above aside .

those are just algorithm for monocular camera . if you want a range sensor , particularly (visually!) stereo camera , you can search for algorithm like KinectFusion .

let me know if these information helps you .