You'll have many options for that, but afaik no working solution out of the box. You can e.g. use scan matching with your point clouds to the known volumetric octomap model, but that's quite involved and computationally expensive.
I did some 3D localization with laser, IMU, and odometry on a humanoid in a known OctoMap model, you can find details in the paper "Humanoid Robot Localization in Complex Indoor Environments". As measurement model, you can immediately apply the ray casting functionality in octomap ("castRay" in OcTree) although I would suggest exploiting parallelization with OpenMP and downsampling the full point cloud.
Edit: I just published the localization code at http://ros.org/wiki/humanoid_localization
It's mostly designated for humanoid robots and still being polished right now, but I'm sure you can use much of the sensor model code as example.