I can't see infrastructure based train detection technology disappearing entirely in the foreseeable future under ETCS L3 or any other moving block system. Train length determination is easy to do for a metro type operation with fixed train lengths, but is much more challenging on a multi-traffic railway which may have to cater for single car passenger units as well as variable length freights. Even on London's LUL and DLR, where train lengths are fixed, The Seltrac moving block system employs axle counter deadlocking through junctions for additional security, as there is little worse in in the rail-verse than a set of points moving under a speeding train whose length has been incorrectly determined, and operators have to cater for all scenarios, including partial failures, unfitted engineering vehicles etc. Those deadlock zones are in fact fully fledged conventional interlockings with detection boundaries split for parallel running of multiple trains through a junction (where safe), sectional route locking to release parts of more complex junctions earlier, and flank protection rules applied as with any conventionally signalled railway. Because they very accurately determine precisely when a train clears a conflict zone, a fixed train detection system can release a junction for another conflicting train far quicker than a calculated offset from the front cab position because of the cycle time in communicating and processing and the uncertainty in train position leads to a generous added tolerance value being neccessary, even with fixed or known train length. On TfL the same wheel counting sensors are also employed for plain line axle counter sections between junctions and stations for degraded mode working with fixed blocks, if train or track equipment fails (albeit with severely reduced capacity). TfL also plans to use Seltrac on the resignalled subsurface network (Circle / District / Hammersmith & City / Metropolitan Lines) in a similar configuration. A new development which may come about in this latest application is to replace the unloved continuous 'wiggly wire' feature (used for position determination) with passive beacons (balises) installed wherever a wire transposition would occur and overlay a full coverage radio system based on COTS (commercial off the shelf) wi-fi technology.
https://www.thalesgroup.com/sites/default/files/asset/document/SelTracBrochure_CBTCSolutions_eng.pdf
A simple single line of railway equipped with a notional moving block system has to be set for a particular 'current of direction' just like a conventionally signalled line with multiple headway signals, then can admit a number of following trains in the same direction whose separation is controlled by other means. At it's simplest such other means could be the drive on sight rules used on light railways such as Manchester Metrolink, but could equally be a sophisticated system of train to train or track to train communication to safely separate faster or more diverse traffic such as ETCS L3. Either way the section needs to be proved clear of vehicles before it's direction can be reversed, just like a conventionally signalled fixed block line.
One idea I had for variable length determination exploits those short deadlocking train detection sections that are likely to remain through junctions. If the sections only admitted one train at a time (not overly restrictive as they are usually short compared to braking distance), then at the moment of their clearing at a particular extremity, the system could compare the known position of the active front cab with the fixed position of the deadlocking sensor and thus determine the train length accurately. This measured value could then be used for following train separation throughout the subsequent moving block section.