Ideally the source code should be littered with ANSI-C assertions.
These are indispensable in detecting segmentation faults due to
dereferencing null pointers.
Obviously there are not enough... so consider adding more!
One way to track down the source of a random segmentation fault, is to litter the pdf2htmlEX and/or poppler source code with printfs... This will allow you to detect the last working code, and this might suggest where to start reading the code in detail.
Poppler uses object reference counting. Unfortunately the assumption about "who" is responsible for incrementing and decrementing these refernce counts does change between poppler releases. This means that between poppler releases the pdf2htmlEX code should include new increments or decrements or now has too many...
Solution: Check all {inc,dec}RefCnt uses... from poppler-0.68.0 the USER (i.e. pdf2htmlEX) is supposed to incRefCnt as well as decRefCnt.