Inspiration## The Inspiration
I became interested in building AVIS after thinking about how difficult it can be to continuously observe wildlife without having a person present all the time. A camera can capture an image, but a useful monitoring system needs to do much more: detect movement, preserve observations, understand environmental conditions, and process the collected images.
This led me to develop AVIS — Autonomous Visual & Environmental Integrated System, a project combining embedded electronics, environmental sensing, cloud technology, and AI-assisted image analysis.
How I Built It
AVIS is built around an ESP32-CAM with an OV3660 camera and microSD storage. I developed a workflow in which the camera detects visual changes, captures an image, validates the JPEG, and saves it to the SD card before attempting to transmit it.
This local-first approach became one of the most important design decisions in the project. If Wi-Fi, a server, or cloud synchronization fails, the original image remains safely stored and can be processed or synchronized later.
I also integrated a DHT22 temperature/humidity sensor and rain sensor to collect environmental information alongside visual observations. The system includes a solar-power subsystem intended to support longer autonomous operation.
On the software side, I developed the image-storage and synchronization workflow and connected the captured images to an AI-based analysis pipeline. During development, the system accumulated more than 4,000 captured images, with 3,666 records processed through the AI-analysis database.
What I Learned
AVIS taught me that building a real-world system is very different from making individual components work separately. Camera capture, image storage, networking, cloud synchronization, sensors, power, and AI all have to work together.
I also learned not to treat an AI confidence score as equivalent to accuracy. For example, one documented House Sparrow prediction had 99.11% model confidence, but this does not mean the system has 99.11% classification accuracy.
Challenges
The project involved many failures and iterations. I encountered problems with camera buffering, corrupted or incomplete JPEG files, unreliable image transmission, network and firewall issues, cloud synchronization, sensor scheduling, and power conversion.
Another important discovery was that motion detection does not automatically mean wildlife detection. Movement from people near the camera also triggered image capture, and some of these non-target images were subsequently passed to the AI system. This showed me that future versions need better wildlife-specific detection before classification.
Rather than treating these failures as setbacks, I used them to redesign and improve the system. AVIS is still an evolving prototype, and my next goal is to conduct more controlled experiments to measure its reliability, AI performance, environmental monitoring, and long-duration autonomous operation.
Log in or sign up for Devpost to join the conversation.