From echo to distance
Sensor
Outgoing pulse
Reflected sound
Obstacle
Distance = speed of sound × round-trip time ÷ 2
The sound travels to the obstacle and back.
VIDI
Reimagining how people who are blind or have low vision interact with the world
IThe problem
The problem people with vision loss face
01Limited technologyTechnology has advanced rapidly, while the basic design of the traditional white cane has changed little.
02Limited sensory informationA traditional white cane helps detect obstacles at ground level and provides limited information about the wider surroundings.
03No warnings about overhead hazardsOverhead hazards such as low branches and open cabinet doors can sit beyond the reach of a traditional white cane.
IIOur solution
Meet VIDI
Visual information, on your terms.
What VIDI does
One frame. Two ways to sense the scene.
- 01Measure distance in the dark. Two ultrasonic sensors measure the distance to nearby obstacles in complete darkness and require minimal computation.
- 02Turn visual information into speech. A camera captures the scene so VIDI can classify objects, recognize text, and speak useful information aloud.
- 03Complement everyday mobility tools. VIDI is designed to work alongside a traditional white cane or guide dog, giving blind and low-vision users more information about their surroundings and helping with everyday tasks.
We bench-tested the sensors and confirmed that the Pi detects the camera.
Features
Object Detection & Recognition | Conversational AI | Scene Description | Text Recognition
IIIEngineering and design
Engineering and design
A modular frame built to be rebuilt.
Mechanical design
Modular, 3D-printed, and serviceable
VIDI uses a modular, 3D-printed glasses frame that integrates multiple sensing and processing components. Reinforced, ergonomic temple arms carry curved over-ear supports, removable electronics mounts, and adjustable ultrasonic sensor holders — so any component can be replaced or modified without redesigning the entire frame.
Hardware schematic · not a product photo
Annotated frame schematic · CAD renderings and prototype photography slot in here
Hardware architecture
- Raspberry Pi 5Main computer for camera processing, AI integration, and speech interaction — housed in its original enclosure with the internal cooling fan.
- ESP32The microcontroller collects distance measurements from two HC-SR04 ultrasonic sensors.
- Dual ultrasonic sensorsMounted on opposite sides of the frame for independent left and right proximity measurements.
- IMX708 cameraCentrally mounted, rotated 90° so its 7.5-inch ribbon cable routes toward the right temple while the lens stays forward-facing.
- Bluetooth headsetHands-free voice interaction and audible feedback.
ESP32 → Raspberry Pi 5 · USB serial · 115,200 baud
Design evolution
Five iterations, five problems solved.
- 01Redesigned the ultrasonic mounts after the original supports obstructed sensor installation.
- 02Reinforced the temple arms to carry the electronics.
- 03Separated the ear-contact structure from the external electronics supports to improve wearability.
- 04Rotated the camera board to work within the limited ribbon-cable length.
- 05Developed a ventilated Raspberry Pi enclosure mount while retaining its cooling system.
Innovation & limitations
VIDI combines camera-based visual interpretation with independently acquired ultrasonic proximity measurements. The modular design emphasizes accessibility, adaptability, and iterative prototyping — and is intended to complement traditional mobility aids, not replace them.
- Electronics weight distribution
- Thermal management
- Sensor blind spots
- Wiring durability
- Real-world validation still ahead
Bench-tested functions are labeled as such; proposed capabilities stay marked as unverified.
IVHow it works
Sense. Process. Speak.
What the ESP32 does
0 ms · Left trigger
Play runs at 12.5 fused packets per second. Step advances 40 ms. Pulse widths are illustrative.
One sensor fires every 40 ms · Each sensor updates every 80 ms · 12.5 Hz per sensor.
Filter echoes
A median filter rejects brief spikes. A tracker uses the filtered readings to estimate distance and closing speed.
Locate the obstacle
The ESP32 aligns both readings in time and uses the sensor spacing to estimate direction. If the readings differ too much, it reports the nearer side.
Estimate time to contact
A distance of 1 m and a closing speed of 0.5 m/s give about 2 seconds to contact. Distance and time to contact together determine the hazard level.
From the ESP32 to the Raspberry Pi
Left and right sensors
ESP32
USB serial
Raspberry Pi 5
JSON over USB serial · 115,200 baud.
The ESP32 processes ultrasonic readings and sends usable obstacle estimates to the Pi.
Example obstacle estimate · abbreviated ESP32 packet
{
"l": 120,
"r": 122,
"brg": -8.2,
"haz": "caution"
}Vision & voice
- 01IMX708 camera
- 02Raspberry Pi 5
- 03Gemini online
- 04Bluetooth headset
VIMeet the team
Meet the team
The five people behind VIDI.
Headshots, majors, and LinkedIn links in place
