New Bullet Impact Dataset Supports Machine Learning Trajectory Reconstruction

Reconstructing ballistic trajectories at crime scenes remains a major challenge in shooting incident investigation.

Because inserting rods relies heavily on personal judgment and differs from one specialist to another, traditional techniques like ballistic rods depend heavily on the individual operator.

Although artificial intelligence presents a way to assist specialists in determining projectile angles, advancement is hindered by a scarcity of accessible public collections containing bullet strikes labeled with their matching firing angles.

As noted by Nature, this study presents a noun database of bullet impact images captured at multiple shooting angles, specifically designed to enable the development and training of machine learning models for projectile trajectory reconstruction. Working alongside the Criminalistics Service of the Spanish Guardia Civil—whose specialized knowledge proved crucial for gathering authentic impact photos—allowed researchers to construct the database.

As technical validation, convolutional neural networks were used to classify impacts by firing angle. According to edatos.consorciomadrono.es, the generated forecasts demonstrate that the technique can determine crucial factors including the shooter’s location and separation from the target area, highlighting its usefulness for reconstructing shooting events.

Traditional techniques like ballistic rods rely on manual insertion, which introduces subjectivity and causes variation between different experts.

The database was built in collaboration with the Criminalistics Service of the Spanish Guardia Civil, using realistic impact images captured at multiple shooting angles to train machine learning models.

Founded in 1844 under Queen Isabel II by the 2nd Duke of Ahumada, the Spanish Guardia Civil was modelled on the French National Gendarmerie to suppress endemic highway banditry and rural unrest. Its first academy opened in Valdemoro in 1855, where graduates received the distinctive tricorne hat still associated with the corps.

As a military-force gendarmerie, it operates from casas cuartel — combined residential garrisons and police stations — and patrols rural zones, highways and ports under the dual authority of the Ministry of the Interior and the Ministry of Defence.

Researchers used convolutional neural networks to classify the impacts by firing angle, demonstrating that the model can estimate the shooter’s position and distance to the impact surface.