In particular I'm interested in any algorithms that can separate multiple tracks from one another reliably.
Historically the algorithms used for understanding tracks in cloud and bubble chambers all ran on human platforms. Interestingly after the first generation they were mostly not physicists, but hired help (and mostly women).
There are some drawbacks of handscanning, however, notably the slow processing speed (even when highly parallelized) and a difficulty in accurately evaluating efficiencies in the analysis.
Recently two interesting things have happened
LArTPCs have been described as "electronic bubble chambers"1 because they potentially provide sub-millimeter level spacial resolution in three dimensions over very large volumes2 and support continuous data acquisition.
I know of two LArTPCs that have led to papers: ICARUS (a large scale device at Gran Sasso) and Argoneut (a small testbed at Fermilab). And there are several either under construction or planned: microBooNE (at Fermilab), the LBNE far detector system (very large future project), and a recently planned near detector for a new beamline at CERN.
As far as I know no one has a really satisfactory, fully machine-driven analysis for large and complicated events as yet. The ICARUS papers I have read all seem to describe results that are partly hand-scanned.While Argoneut has published a number of papers, work continue on improving the analyzer.
The basis of the software that we are writing for Argoneut, microBooNE and LBNE is cobbling together a number of well known tools from the image recognition community (clustering, line finding, vertex finding) with some old workhorses from particle physics (Kalman filters, outside in approach), and we can get good results for some events. Work continues on several lines of attack.
1 This is a little bit optimistic because there are some track configurations that could be seen in a real bubble chamber that will baffle any conceivable analysis in a TPC, but it is believed (and hoped!) that fluxuations in the tracks will break the troublesome symmetry in almost all cases.
2 The one I'm working on is more than 50 cubic meters of active volume and the LBNE far detector is planned to be many times that size.
Full disclosure I am named as an author on some Argoneut papers, and am a collaborator on microBooNE and LBNE.