KalyeSuri: Tricycle and delivery-rider phone road-hazard mapper linked to DPWH contractor warranties

Problem Statement

Unmarked potholes, shattered concrete reblocking slabs, sunken drainage grates, and erased school-zone crosswalks cause thousands of motorcycle, tricycle, and bicycle crashes across the Philippines every year. Citizens frequently complain that freshly paved roads crumble after a single rainy season, yet ordinary commuters have no way of knowing that under Section 62 of the Government Procurement Reform Act (RA 9184), road contractors are bound by a mandatory 1-year Defects Liability Period (and up to 5–15 years structural warranty) during which the contractor must repair defects at their own expense, not taxpayers’ expense. Because manual road inspections are rare and citizens don’t know which road kilometer belongs to which contractor, defective roads either stay dangerous for months or get re-budgeted twice.

Proposed Solution

KalyeSuri turns everyday tricycle drivers (TODA), motorcycle delivery riders, and cyclists into an automated, zero-effort road safety and public-works warranty audit network:

1. Passive Handlebar Vision + IMU Jolt Sensing: A rider places their phone in their standard handlebar navigation mount. Whenever the phone’s 3-axis accelerometer/gyroscope detects a sharp vertical jolt spike ($> 2.2g$ on the Z-axis) or at 2-second intervals along a route, KalyeSuri captures a road-surface frame and runs a quantized YOLOv8n model trained on the multinational RDD2022 benchmark to classify 4 standard defect types: D00 Longitudinal Cracks, D10 Transverse Cracks, D20 Alligator Cracks, D40 Potholes / Sunken Manholes, and Faded Crosswalks.
2. Multi-Rider Consensus Verification: When 3 or more riders independently trigger a vision-plus-accelerometer hit at the same 10-meter GPS road segment, KalyeSuri promotes the pin to a Verified Road Hazard, eliminating single-phone false positives.
3. Automated DPWH / LGU Contractor Warranty Match: KalyeSuri spatially joins the defect’s GPS coordinate and road kilometer stationing against published DPWH PCMA (Project and Contract Management Application) and LGU road-reblocking contract records. If the road was completed within the last 12 months, it flags the pin in red as Under Contractor Warranty — Free Repair Mandatory (RA 9184 Sec. 62), names the contractor and contract ID, and generates a one-click demand letter for the Barangay Council, City Engineer, and COA Citizen Participatory Audit desk.

Target Market

• Primary users: Tricycle Operators and Drivers’ Associations (TODA), motorcycle delivery riders, bicycle commuter groups, and school parent-teacher associations (PTAs).
• Public-sector and civic buyers: City and Municipal Engineering Offices, DPWH District Engineering Offices, Commission on Audit (COA) Citizen Participatory Audit teams, and road-safety NGOs.
• Community benefit: Prevents fatal two-wheeler crashes while stopping taxpayers from paying twice for substandard road projects.

How AI Is Used

• Road damage object detection (RDD2022 YOLOv8n): Detects and bounds potholes, alligator cracking, and shattered slab joints in moving dashcam/handlebar video frames while blurring vehicle license plates and pedestrian faces on-device.
• IMU accelerometer + vision sensor fusion: Combines vertical G-force shock waves from the phone’s inertial measurement unit with visual bounding-box area to score physical pothole depth and crash severity for two-wheelers.
• Spatial linear-referencing matcher: Maps raw GPS points onto DPWH kilometer-post stationing (K0014+200 to K0014+850) to link physical defects directly to itemized government contract records.

Community & Ecosystem Integration

• Enforces RA 9184 Section 62 (Contractor Defects Liability): Shifts community reporting from “begging for a repair budget” to enforcing existing contractor warranties at zero cost to the LGU.
• TODA & Rider Hazard Map: Gives night-shift riders and tricycle drivers an audio voice warning (“Babala: malalim na lubak sa 50 metro”) when approaching newly reported potholes during rain or poor street lighting.
• Zero manual typing while riding: Runs completely hands-free in the background while the rider works.

Demo Scenario (3 minutes for judges)

1. Play a 20-second handlebar video recorded along a Philippine city street with potholes and cracked concrete slabs while shaking a connected phone to simulate the physical wheel jolt.
2. Live on the screen, KalyeSuri draws red bounding boxes around two deep potholes (D40 Pothole — Confidence 94%) synchronized with a 2.8g Z-axis accelerometer spike, while automatically blurring passing license plates.
3. The defect pin lands on the PostGIS map at Kilometer Station K0018+340.
4. KalyeSuri auto-matches the segment to a sample DPWH road-concreting contract completed 7 months ago (Contract ID 25O00123 — ₱18.4M).
5. A red banner pops up: ACTIVE DEFECTS LIABILITY PERIOD (5 months remaining) — Contractor Must Repair at Zero Public Cost. Click Generate COA / District Engineer Notice to export the photo-and-GPS-stamped audit dossier.

30-Hour Build Plan

Suggested stack: Python FastAPI, Ultralytics YOLOv8 (pretrained on RDD2022), Web DeviceMotion API / Flutter Sensors, PostgreSQL/PostGIS, Next.js + MapLibre GL. Public data / APIs: RDD2022 (Multi-National Road Damage Detection Dataset), DPWH public contract award notices and road atlas kilometer stationing, OpenStreetMap Philippines road network.

Research & Literature References

1. Deeksha Arya, Hiroya Maeda, Sanjay Kumar Ghosh, Durga Toshniwal, and Yoshihide Sekimoto, “RDD2022: A multi-national image dataset for automatic Road Damage Detection,” Geoscience Data Journal, Vol. 11, Issue 4, pp. 846-862, 2024. DOI: 10.1002/gdj3.260
   • How KalyeSuri uses this study: Uses the RDD2022 standardized damage taxonomy (D00, D10, D20, D40) and smartphone-mounted road inspection benchmarks to detect potholes and structural pavement cracking from low-cost handlebar cameras.
2. Deeksha Arya et al., “Crowdsensing-based Road Damage Detection Challenge (CRDDC’2022),” 2022 IEEE International Conference on Big Data (Big Data), pp. 6378-6386, 2022. DOI: 10.1109/BigData55660.2022.10021040
   • How KalyeSuri uses this study: Adopts the crowdsensing mobile-edge architecture evaluated in CRDDC’2022 to aggregate road defect detections across everyday vehicles at city scale.

Risks & Privacy Notes

• Rider safety (zero screen distraction): The capture mode has no buttons to press while moving; the screen stays dark or shows standard navigation while jolt + vision triggers run passively.
• Automatic bystander and plate anonymization (RA 10173): All pedestrian faces and vehicle license plates in road frames are blurred before any defect image is saved to the public map.
• Multi-report verification: Requires either 3 independent rider hits or a clear vision + IMU match before generating an official contractor warranty notice.
• Strict 30-hour feasibility constraint: DPWH Project and Contract Management Application (PCMA) kilometer-stationing boundaries are not exposed via a clean public API and must be manually digitized per highway corridor.

AI Use Disclosure (draft)

KalyeSuri uses a YOLOv8 object detection model trained on the RDD2022 benchmark to identify road surface defects, inertial sensor fusion to measure jolt severity, and an LLM to draft formal RA 9184 warranty repair requests. AI coding assistants helped write the PostGIS linear-referencing queries and map dashboard. All contract warranty rules follow the Implementing Rules and Regulations of RA 9184.

Rubric Scoring