Low power CMOS vision sensor for foreground segmentation

• Autores: Daniel García Lesta
• Directores de la Tesis: Paula López Martínez (dir. tes.), Victor Manuel Brea Sánchez (dir. tes.)
• Lectura: En la Universidade de Santiago de Compostela ( España ) en 2021
• Idioma: inglés
• Tribunal Calificador de la Tesis: Jorge Fernández-Berni (presid.), Natalia Seoane Iglesias (secret.), Mika Laiho (voc.)
• Programa de doctorado: Programa de Doctorado en Investigación en Tecnologías de la Información por la Universidad de A Coruña y la Universidad de Santiago de Compostela
• Materias:
  • Matemáticas
    • Análisis numérico
      • Construcción de algoritmos
  • Física
    • Electrónica
      • Circuitos integrados
• Enlaces
  • Tesis en acceso abierto en: MINERVA
• Resumen

  • This thesis focuses on the design of a top-ranked algorithm for background subtraction, the Pixel Adaptive Based Segmenter (PBAS), for its mapping onto a CMOS vision sensor on the focal plane processing. The redesign of PBAS into its hardware oriented version, HO-PBAS, has led to a less number of memories per pixel, along with a simpler overall model, yet, resulting in an acceptable loss of accuracy with respect to its counterpart on CPU. This thesis features two CMOS vision sensors. The first one, HOPBAS1K, has laid out a 24 x 56 pixel array onto a miniasic chip in standard 180 nm CMOS technology. The second one, HOPBAS10K, features an array of 98 x 98 pixels in standard 180 nm CMOS technology too. The second chip fixes some issues found in the first chip, and provides good hardware and background performance metrics.