Background of the Problem Statement
Government-funded skilling schemes rely on a large, geographically dispersed network of empaneled training centres, many of which are physically inspected only periodically. This creates two recurring integrity risks: attendance records that do not reflect actual physical presence of trainees, and centres whose approved infrastructure (equipment, seating capacity, workshop facilities) is not consistently available or maintained between inspection cycles. Manual, periodic inspection is resource-intensive and cannot catch issues that arise between visits, and existing camera infrastructure at many centres is used only for passive recording rather than active compliance monitoring. An AI-based system that can process centre camera feeds in near real time - flagging attendance discrepancies and infrastructure gaps as they occur rather than only at the next scheduled audit would materially reduce leakage and improve the credibility of scheme delivery data reported up to the ministry.
Description of the Problem Statement
The challenge is to build an AI-based video analytics system that can:
Process live or periodic camera feeds from training centres to estimate actual attendance and cross-check it against records submitted by the centre. Detect the presence/absence and apparent operability of approved infrastructure items(workbenches, machinery, seating) against the centre's sanctioned inventory. Flag discrepancies (attendance mismatches,missing/non-functional equipment) to a monitoring dashboard for follow-up, rather than requiring manual review of raw footage. Operate within realistic bandwidth and camera-quality constraints found at rural and semi-urban training centres. Preserve trainee privacy using aggregate presence-detection rather than facial identification wherever a compliance check does not require individual identification.
Expected Solutions / Outcomes
A working video-analytics pipeline demonstrated on sample/simulated centre footage. An attendance-discrepancy and infrastructure-compliance dashboard for scheme monitoring units. A privacy-preserving design note explaining what is and isn't identified from footage. A false-positive/false-negative accuracy assessment on the demonstration dataset. A low-bandwidth deployment mode suitable for centres with limited connectivity.