Autonomous Parking Sign AI: YOLOv11,

Autonomous Parking Sign AI: YOLOv11, OCR & Deterministic Rule Engine.

The primary objective of this project was to architect an end-to-end computer vision and reasoning pipeline, named CurbAI, capable of translating complex, real-world parking sign imagery into structured, curb-specific legal decisions. Focused on the intricate municipal regulations of Toronto, the system was designed to solve the inherently confusing nature of urban parking rules by systematically decoding overlapping signs, time windows, and spatial constraints to determine definitive legal parking statuses.

To ensure high accuracy and scalability, the system architecture explicitly decouples the visual perception layer from the underlying business logic. The perception pipeline is powered by a robust two-stage YOLO framework. The first stage detects the entire signboard assembly in the wild, while the second stage precisely isolates individual sign plates. These strategically cropped regions are then fed directly into Google Cloud Vision OCR for highly accurate text extraction and advanced token normalization.

Once the visual data is converted into normalized text, it is ingested by a custom-built deterministic rule engine. This sophisticated parsing layer evaluates the extracted text tokens—including specific time windows, days of the week, directional arrows, permit codes, and complex inverse clauses such as "At Other Times." The engine systematically translates these disparate pieces of visual information into formal, machine-readable logical constraints.

The final stage of the pipeline is the Arbitration and Decision Layer, which handles the complex task of resolving overlapping and directly conflicting parking restrictions. By applying a programmed municipal hierarchy of prohibitions, the engine intelligently arbitrates these conflicts to determine the final, authoritative legal parking status. This status is calculated and outputted entirely independently for both the left and right sides of the curb, resulting in a completely automated and highly reliable legal decision framework.