||In April 2010, the eruption of Eyjafjallajökull (Iceland) threw volcanic ash across northwest Europe for six days which led to air travel disruption. This recent crisis spotlighted the necessity to parameterise plume dynamics through emission, dispersion and fall out as to better model, track and forecast cloud motions. This eruption was labeled as a Strombolian-to-Sub-Plinian eruption type. Strombolian eruptions are coupled with a large range of volcanic event types (Lava flows, paroxysms) and eruption styles (Hawaiian, Sub-plinian) and offer a partial precursory-indicator of more dangerous eruptions. In addition, strombolian eruptions are small enough to allow observations from within few hundred meters with relative safety, for both operators and material. Since 2001, thermal cameras have been increasingly used to track, parameterise and understand dynamic volcanic events. However, analyses, modelling and post-processing of thermal data are still not fully automated. In this study, we use thermal video data to fully parameterise the emission dynamics of bombs, blocks, and lapilli. Our aim is to capture the vent leaving properties, as close to the point of fragmentation as possible, of all measurable particles, primarily size, shape and mass. I developed an algorithm to detect pyroclastic particles, track them during their ascent, and extract particle parameters. The algorithm was validated by experimental data, during which I replicated particles with hot ball-bearings. Then, I applied the algorithm on 31 eruptions recorded during two field-trips to Stromboli in 2012 and 2014. The algorithm was able to extract parameters for 83 000 particles generating a statistically robust database. Finally, these output strengthen support for a model suggestive of the presence of a cap at the head of the magma column during pyroclastic-dominated explosions, and its absence during gas-dominated events.