SIH26166Strong pickacceptance 4/5

Multi-modal, Sun angle and scale invariant image correspondence using Chandrayaan-2 optical images (OHRC, TMC and IIRS)

Indian Space Research Organisation(ISRO) · Space Technology · Software

Well-specified with public data and named metrics, and learned matchers give a real path — build for the sun-angle and cross-sensor cases specifically, because those are where the sub-pixel bar bites and where an ISRO judge will look first.

Data: Chandrayaan-2 OHRC, TMC-2, IIRS lunar imagery (link TBD)

What it actually is

Registering two lunar images — aligning them into one coordinate system — is hard when they were shot at different sun angles, from different viewpoints, at different scales, or by different sensors. The ask is a generic tool that finds matching points between Chandrayaan-2 optical images and a lunar reference and aligns them to sub-pixel accuracy, robust to all three of those variations.

What to build

An image-registration pipeline that takes a Chandrayaan-2 optical image and a lunar reference image and finds correspondence points robust to illumination variation from changing sun azimuth and elevation, viewpoint distortion, scale differences between payloads at different altitudes, and cross-sensor differences between OHRC, TMC-2 and IIRS, then computes a geometric transform aligning source to reference to sub-pixel accuracy with match points distributed uniformly across the image rather than clustered, reporting evaluation metrics such as RMSE, inlier match count and inlier ratio.

Smallest thing that wins the room

Take two lunar images of the same region captured at very different sun angles and scales, show your pipeline finding well-distributed correspondence points where classical feature matching fails, and align them with the reported sub-pixel RMSE and inlier ratio.

How crowded this one gets

A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.

Moderate90–210 teams expectedroughly 1 in 76–176 wins it

Quieter than 70% of the 226 · #68 of 226 by expected field

A normal-sized field. Your idea has to be good, not miraculous.

Why: defence, intelligence and space bodies drew small fields.

This is a guess, not a fact

Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.

The scores

The number is the shorthand. The line under it is the reason.

Acceptance potential

4/5

A strong pick — the data is public, the requirement is precise with named metrics, learned matchers give you a real path, and the lunar-imaging domain thins the field, though sub-pixel accuracy under extreme illumination is a genuine bar and cross-sensor matching is the hard case a knowledgeable ISRO judge will test.

Feasibility

3/5

Chandrayaan-2 imagery is publicly available through ISRO's data portal and learned feature matchers like SuperGlue and LoFTR handle exactly this kind of illumination and viewpoint invariance, but sub-pixel accuracy on the lunar surface under extreme sun-angle change is genuinely hard, and cross-sensor matching between an optical camera and a hyperspectral imager is harder still.

Innovation scope

4/5

Making correspondence robust simultaneously to illumination, viewpoint, scale and sensor modality on lunar terrain, with uniform match distribution and sub-pixel accuracy, is genuinely open — off-the-shelf matchers do not solve the sun-angle and cross-sensor cases well.

Clarity

5/5

The description defines the registration components, names the three challenge types precisely, specifies the payloads, states the sub-pixel accuracy and uniform-distribution requirement, and names the evaluation metrics, making it exceptionally well-specified.

Effort

Heavy

Building and adapting a robust matcher to lunar imagery, handling cross-sensor and multi-scale cases, and validating to sub-pixel accuracy is focused, demanding computer-vision work.

Demo-ability

Medium

A registration that succeeds where classical matching fails is a clear result, but it is quantitative — RMSE and inlier ratios — rather than visually dramatic, so the improvement needs the metrics to land.

In its favour

Green flag: Chandrayaan-2 imagery is publicly available through ISRO's data portal, so you can develop on the real target data
Green flag: Learned matchers like LoFTR are built for the illumination and viewpoint invariance this problem needs, giving you a genuine head start
Green flag: The named metrics — RMSE, inlier ratio — mean your evaluation is objective and pre-defined rather than argued
Green flag: The lunar and cross-sensor specialisation guarantees a thin field of serious competitors

Against it

Red flag: Sub-pixel accuracy under extreme sun-angle change is a hard bar, and classical and even learned matchers degrade exactly where illumination differs most
Red flag: Cross-sensor matching between optical and hyperspectral IIRS imagery is the hardest case and the one most likely to be tested
Red flag: The uniform-distribution requirement rules out solutions that only match the few high-texture regions, which is where matchers naturally concentrate
Red flag: An ISRO judge works with this data professionally and will probe the failure cases a demo on easy pairs would hide

What you will be writing

Learned feature matching (SuperGlue / LoFTR)
Chandrayaan-2 OHRC / TMC-2 / IIRS imagery
Illumination-invariant descriptors
RANSAC + sub-pixel refinement
Cross-modal (optical-hyperspectral) registration
RMSE / inlier-ratio evaluation
Planetary remote sensing
Image registration
Computer vision

Prior art to read before you start

illumination-invariant image correspondence · multi-sensor lunar image registration · sub-pixel geometric alignment

Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.