🧍♂️📡 Real-time person detection on Raspberry Pi 5! HitoMi-Cam uses clothing spectra, not shape. Robust to atypical postures where CNNs fail. Ideal for edge-based search & rescue.
by Shuji Ono
🔗 https://t.co/YH1FJ4CGA3
#MDPIjimaging#EdgeAI#ComputerVision#SpectralImaging
🏔️ JIC 2025 concluded after 9 years
I wrote about our "HitoMi-Cam" concept—
a camera that sees "clothing," not "shape"
📝 LinkedIn: https://t.co/cBDImtjBP0
📄 Paper: https://t.co/FPu8wHEOzg
#JIC2025#HitoMiCam#SAR#Robotics
It's not a CNN replacement, it's a complementary tool.
And it's fast: 23.2 fps on a Raspberry Pi 5 (no GPU!).
Watch the live demo vs. CNNs:
📺 Demo: https://t.co/jQQy6R6Nzr
#EdgeAI#RaspberryPi
I'm excited to share my new paper in J. Imaging! We all use CNNs for person detection, but what happens when the target is fallen, occluded, or doesn't look like a "person"?
They fail. This is a critical gap, especially in Search and Rescue (SAR).
Paper: https://t.co/FPu8wHEOzg
HitoMi-Cam is shape-agnostic.
It doesn't see "shape"; it sees the spectral signature of clothing material.
In simulated SAR tests, HitoMi-Cam achieved 93.5% AP. The best-performing CNN? Only 53.8%.
#SpectralImaging#PersonDetection
Following up on our video demonstration, here is the link to the research paper on HitoMi-Cam. Discover the methodology that sets it apart. https://t.co/63D2HouDEr
#HitoMiCam#ObjectDetection#AI#ResearchPaper#Tech
Ever wondered what makes an object detection model truly effective? Our work on HitoMi-Cam, which outperforms models like MobileNet V1 and YOLOv5n, is detailed in our latest publication. Dive into the science behind the performance. https://t.co/63D2Hou5OT
#AI#ComputerVision