Context

Fusce sed enim gravida, pharetra lectus lobortis, consequat metus. Pellentesque vulputate odio nec erat congue, a tincidunt orci ornare. Nunc vehicula quis est vitae hendrerit. Phasellus quis fermentum est, ut aliquam tortor. Integer at est diam. Nunc maximus cursus euismod. Nulla et tellus non nibh aliquet lacinia. Nulla a mauris ac metus aliquet vestibulum sed eu metus. In sollicitudin ut leo vitae tincidunt. Donec faucibus metus efficitur, facilisis dolor a, scelerisque urna. Vivamus sapien odio, pellentesque sit amet sagittis vel, rutrum vel nisi. Duis ac ultrices urna, non auctor neque. Nunc sed risus tempus, dictum nulla a, accumsan orci. Vivamus condimentum turpis efficitur efficitur tempor. Ut in dictum leo. Vivamus sit amet mollis dolor.

The eco-conservation non-governmental organization Nisarggatri actively documents ancient geoglyphs carved into porous laterite rock faces across the Konkan region. These historical carvings are exceptionally shallow and easily obscured by natural elements, particularly during heavy monsoons which trigger intense surface reflectivity and rapid vegetation growth that completely masks the prehistoric art. Operating with highly limited funding and completely lacking the budget for expensive multi-spectral aerial scanning hardware, the organization’s field workers were forced into a severe data-entry bottleneck, spending hundreds of hours manually tracing and extracting rock carving outlines to export the historical site layouts into GIS mapping software.

Requirement

I was brought into a hands-on, 3-day technical discovery workshop at the NGO’s regional research center in Ratnagiri. I was tasked with establishing an optimal, highly cost-effective 3D scanning, data processing, and automated computer vision workflow to accurately isolate, document, and preserve these climate-threatened ancient assets without relying on expensive hardware.

Solution

I began by processing over 200 high-resolution field photographs of the prominent and intricate Chande rhinoceros geoglyph site, converting the image sets into a dense, geocoded 3D point cloud. Based on this fieldwork, I developed a custom photogrammetry Standard Operating Procedure (SOP), establishing optimal camera angles and lighting balances specifically calibrated to counteract rock face reflectivity and shallow carving depths for future field workers. To help the research center generate self-sustaining educational materials and souvenirs, I engineered a modular, two-piece, multi-colored snap-fit 3D print model directly from the raw 3D scan data, a design choice that sharply cut traditional production time and material costs down to a fraction of standard baselines. Finally, to eliminate the manual data-entry bottleneck, I proposed a computer vision solution using Meta’s open-source Segment Anything Model 2 (SAM 2) foundation framework. I manually masked a precise, custom training dataset of roughly 100 historical site images and used it to fine-tune a specialized model checkpoint, successfully automating the process of isolating shallow carving boundaries from the surrounding porous rock textures.

Outcome

The engagement provided the conservation group with an immediate, actionable technology roadmap, combining immediate field-ready photogrammetry workflows, an affordable merchandising blueprint, and structural frameworks for future interactive virtual reality educational exhibits. The fine-tuned computer vision prototype successfully proved that targeted, low-cost open-source AI models could automate spatial documentation, saving the NGO hundreds of hours of manual asset tracing, bypassing the need for expensive multi-spectral hardware, and drastically accelerating the cultural heritage mapping pipeline.