The heart of the system is remarkably familiar hardware. Kosmiko is built around a Raspberry Pi 5 single-board computer that simultaneously controls two Sony IMX477 camera modules, the same class of sensor found in the popular Raspberry Pi High Quality Camera. The researchers operate these complementary metal-oxide-semiconductor, or CMOS, active-pixel sensors in raw acquisition mode with the lenses capped, meaning the cameras work in complete darkness. In that darkness, the only signals that should appear are electronic noise and, crucially, the faint electrical fingerprints left behind when an ionizing particle passes through the silicon. Each crossing particle generates a shower of electron-hole pairs along its track, and the sensor readout electronics record this deposited charge as an unusually bright or saturated pixel, or as an elongated cluster of pixels tracing the particle’s path.

What separates Kosmiko from a simple curiosity is the rigor of its calibration strategy. CMOS sensors are imperfect devices: individual pixels differ in their dark response, some run persistently hot, others are noisy, and a few fire spuriously regardless of conditions. Rather than treating these defects as a nuisance, the Montpellier team places each pixel on its own footing. Every measurement run begins with a calibration step in which the intrinsic threshold of each pixel is derived from its statistical behavior in darkness. During the subsequent continuous image acquisition, the intensity of each pixel is compared online against its own personalized threshold. Only pixels whose intensities exceed their individual thresholds are stored; everything else is discarded immediately. This per-pixel discrimination drastically reduces the data volume, which is what makes long-duration acquisitions practical on modest, low-power hardware.

The methodology also confronts the problem of bad actors among the pixels. Hot pixels, noisy pixels, and recurrent pixels that repeatedly exceed their thresholds for mundane reasons are identified and filtered out, so that genuine particle candidates are not drowned in a sea of sensor pathology. The authors complement this with a suite of camera-health indicators based on dark-level statistics, noise distributions, and percentiles, allowing an operator to verify at any time that the sensor is behaving as expected and that any change in event rates reflects the radiation environment rather than a drifting instrument. This kind of self-monitoring is essential for any detector intended to run unattended for days or weeks.

Once the raw frames have been reduced to candidate events, the offline analysis layer of Kosmiko extracts physically meaningful observables. These include the frame-of-interest rate, which counts how many images contain at least one candidate, and the cluster rate, which counts localized groups of bright pixels. The software also characterizes the intensity distributions of events, indicators of track-like topology that suggest a particle crossed the sensor rather than merely triggering a single defective pixel, and the occurrence of saturated pixels, whose maximum-registering response is typical of heavily ionizing particles depositing large amounts of energy. Together these observables allow researchers to compare ionizing-event rates under different conditions without pretending to know more than the data can support.

Indeed, the team is explicit about the scope of the instrument. Kosmiko is not designed to identify individual particle species. A muon streaking down from the upper atmosphere, a gamma ray from a radioactive decay, and an alpha particle from radon progeny all leave charge in silicon, and disentangling them requires more than intensity histograms alone. Instead, the goal is a documented and deployable methodology for comparing ionizing-event rates under different conditions in the natural radiation environment. That comparative capability is scientifically valuable in its own right, because variations in event rate with altitude, shielding, underground depth, or proximity to weak radioactive sources all carry information about the composition and behavior of ambient radiation.

The experimental evidence presented in the paper spans an impressive range of radiation conditions. The authors performed tests with multi-alpha emitters and with a cobalt-60 gamma source using the PRESERVE platform, a facility funded by the French Alternative Energies and Atomic Energy Commission, the Occitanie region, and the European Union through European Regional Development Funds. They conducted ground-level measurements of the natural background, and, in collaboration with the Laboratoire Souterrain à Bas Bruit in Rustrel, France, a National Underground, Low-Noise Laboratory operated with CNRS, they acquired data deep underground, where the cosmic-ray flux is heavily suppressed by hundreds of meters of rock. Comparing event rates across these radically different environments demonstrates exactly the kind of condition-to-condition sensitivity the instrument was built to provide.

The Kosmiko project sits within a growing movement to democratize particle physics. Previous efforts have shown that smartphone cameras can detect muons, and distributed networks of amateur detectors have been proposed as a complement to large observatories. Studies using CMOS sensors in cell phones for gamma detection and classification, along with dedicated cosmic-ray camera experiments using Raspberry Pi hardware, have established the basic feasibility of the idea. Kosmiko advances this tradition by combining it with the radiation-hardening and reliability expertise of the Montpellier group, whose prior work has focused on how atmospheric neutrons and natural radiation cause single-event upsets in advanced microelectronic memories. The same physics that threatens avionics and satellites, in other words, becomes a measurable signal in a hobby-grade camera.

Reproducibility is the watchword throughout the publication. The complete hardware configuration, the software workflow, and the resulting data products are described in detail so that other laboratories, classrooms, and citizen scientists can rebuild the platform and compare results meaningfully. The analysis pipeline is built on widely used open scientific tools, and the authors discuss practical observations about the camera components and the local measurement environment that would otherwise have to be rediscovered by trial and error. The team notes that no specific funding was received for the work, underscoring how inexpensive the platform is; the write-up credits Frédéric Wrobel with designing the experimental concept and methodology, with colleagues at the Institut d’Électronique et des Systèmes and the underground laboratory contributing to the alpha, cobalt-60, ground-level, and underground campaigns.

The implications reach beyond fundamental physics. A network of calibrated, low-cost pixel detectors could monitor natural radiation environments across wide geographic areas, support education by letting students see particle tracks with their own equipment, and provide empirical grounding for models of the terrestrial cosmic-ray flux such as those used to estimate radiation doses and predict electronic failure rates. Because the entire chain, from consumer sensor to per-pixel calibration to event-rate observables, is documented and deployable, Kosmiko lowers the barrier to participation in radiation measurement to the price of a small computer and two camera modules. In an era when frontier physics often demands billion-euro machines, there is something quietly revolutionary about silicon meant for photography quietly recording the passage of particles born light-years away, one calibrated pixel at a time.

Subject of Research: Low-cost CMOS image sensors used as pixel-calibrated detectors of natural ionizing radiation

Article Title: Kosmiko: from CMOS imaging sensors to pixel-calibrated particle instruments

Article References: Wrobel, F., Wrobel, J., Maraine, T., Boch, J., Risso, C., Dilillo, L., & Saigné, F. (2026). Kosmiko: from CMOS imaging sensors to pixel-calibrated particle instruments. Experimental Astronomy, 62(2), Article 21. https://doi.org/10.1007/s10686-026-10081-9

Image Credits: AI Generated

DOI: 10.1007/s10686-026-10081-9

Keywords: cosmic rays, CMOS sensor, active-pixel sensor, Raspberry Pi, ionizing radiation, Sony IMX477, particle detection, calibration, radiation environment, Experimental Astronomy, citizen science, underground laboratory