Tipus de document
Treball de fi de grauData de publicació
Llicència de publicació
Si us plau utilitzeu sempre aquest identificador per citar o enllaçar aquest document: https://hdl.handle.net/2445/231657
Development of a Minimalist Camera for Lightweight Computer Vision Applications
Títol de la revista
Director/Tutor
ISSN de la revista
Títol del volum
Recurs relacionat
Resum
Traditional computer vision architectures rely on multi-megapixel image sensors that capture dense grids of pixels. While this approach is well-suited for human viewing, it introduces significant data redundancy, high computational load, and substantial energy consumption, particularly unnecessary in lightweight vision tasks such as occupancy sensing, motion detection, or environmental monitoring. Furthermore, continuous high-resolution image capture raises critical privacy issues in shared spaces.
This bachelor’s thesis presents the design, hardware implementation, and experimental characterization of a dedicated acquisition platform for a minimalistic camera (MiniCam) based on masked sensing and the introduction of modern approaches to build masks using deep learning models (free-form pixels). By replacing CMOS pixel grids with an array of 24 discrete silicon photodiodes interfaced by custom optical masks. The compression of the measurements is carried out directly in the optical stage, reducing the amount of visual data captured while maintaining the relevant information.
The developed hardware encompasses a three-stage analog signal-conditioning chain per channel, consisting of a transimpedance amplifier (OPA381), a passive low-pass filter, and a secondary voltage gain stage (MCP6001), connected to three 8-channel, 12-bit SAR analog-to-digital converters (MCP3208).
The acquisition and wireless communication are managed by an ESP32 microcontroller embedded into the so-called “Dolfins platform” designed previously at the CEMIC (UB), which transmits 24-channel measurement packets at 30 frames per second (fps) via Bluetooth Low Energy (BLE) to a customized
graphical interface.
The system was evaluated through an automated lightbox characterization setup across various illumination levels and color temperatures, confirming high linearity (𝑅2 > 0.95) and an effective transmittance resolution of approximately 4 bits under standard indoor lighting. Electrical characterization demonstrated a stable average power consumption of 426.42 mW (129.22 mA at 3.3 V). The resulting platform validates the technical and economic feasibility of minimalist vision hardware, laying the foundation for ultra-low-power, privacy-preserving embedded optical sensing nodes.
Descripció
Treballs Finals de Grau en Enginyeria Electrònica de Telecomunicació. Facultat de Física. Universitat de Barcelona. Curs 2025-2026. Tutor/Director: Ismael Benito Altamirano
Matèries (anglès)
Citació
Citació
MALLADA MARTINEZ, Pablo Salvador. Development of a Minimalist Camera for Lightweight Computer Vision Applications. [consulted: 25 of September of 2026]. Available at: https://hdl.handle.net/2445/231657