Road infrastructure failure is a lethal and expensive problem in India: government figures for 2022 attribute thousands of fatalities nationwide to potholes and road-surface defects, and municipal budgets commit tens of thousands of crore rupees annually to repairs that citizens report are rarely completed on time. Existing tools address isolated slices of this pipeline — automated damage detection, civic reporting applications, or open-data dashboards — but none have integrated detection, repair-cost estimation, fraud-robust citizen ingestion, and public contractor accountability in a single deployable system. We present CrackWatch, an open-source reference implementation that unifies all four. CrackWatch combines a three-model fusion pipeline (YOLOv8s fine-tuned on RDD2022, a dedicated crack-segmentation model, and classical OpenCV heuristics for spalling, leak, and corrosion) with a five-layer fraud detector specifically tuned for crowd-submitted infrastructure photographs, a rule-based monsoon-adjusted decay model for prioritization, and a public "Wall of Shame" ranking in which contractor negligence is computed deterministically from unresolved-report dwell time and severity. The system ingests reports through three channels — a government dashboard, a citizen Progressive Web App, and a Twilio-backed WhatsApp bot — making it usable by the estimated 500 million WhatsApp users in India without application installation. We describe the architecture, detection and scoring algorithms, the accountability model, and a reference deployment. We report wall-clock latency and resource usage, and we explicitly separate substantiated engineering claims from aspirational ones so that the work is reproducible and its limitations are clear. Winning entry, Nirman 2026 Hackathon hosted by Amity University Mumbai. Source code: https://github.com/SaudSatopay/CrackWatch-NirmanHackathon
Satopay et al. (Mon,) studied this question.