Hong Kong's vibrant streetscape, a captivating blend of East and West, is punctuated by a dense tapestry of shopfront signage. These signboards, often elaborate and overflowing with character, tell a story of the city's entrepreneurial spirit and cultural dynamism. They are more than just advertisements; they are integral to the urban fabric, contributing significantly to Hong Kong's unique visual identity. However, this iconic visual landscape faces an often-overlooked challenge: the deterioration of these very signboards. Exposure to Hong Kong's humid climate, typhoons, and the relentless passage of time takes its toll, leading to rust, breakage, and even gone missing. These seemingly minor damages not only detract from the aesthetic beauty of the city but also pose potential safety hazards.
What if using a vision-based system mounted on top of a vehicle roaming around the roadways and urban canyons of Hong Kong automatically captures these defects and report to the authorities and stakeholders timely?
Such defects have been detected at WISIO that can be tested and deployed in real-time. The algorithms take video frames facing building facades from a camera mounted on a car and judge the condition of these signboards whether they have become obsolete or broken or normal. The algorithm also detects if the signboards have been missing and that only the holding frames are remaining. A system is proposed that first detects the signboards, then categorize them, visualize the output, and analyze the performance to continuously improve its performance to the new situations:
The proposed system flowchart
The algorithm is tried on images with varying resolutions and conditions, hence, “in the wild”. Similar to the photos below:
With only 50 scene captures for each category (broken, missing, signs of obsolete, normal) randomly divided into train, validation, and testing sets, two of our preliminary models achieve over 92% accuracies in reporting the signboard current condition.
Looking forward
The development of an automated damage detection system for Hong Kong's shopfront signboards offers an important practical application. This technology can streamline inspection processes, reducing the time and cost associated with manual assessments. By providing cheaper and timely alerts about damaged signs with wider coverage, the system contributes to improved public safety by mitigating potential hazards. Furthermore, the data collected can inform targeted maintenance strategies, optimizing resource allocation and ensuring the long-term preservation of Hong Kong's distinctive visual identity. This project, therefore, represents a valuable contribution to both urban management and the preservation of Hong Kong's unique cultural heritage. The efficient and accurate identification of damaged signage paves the way for a safer, more beautiful, and better-maintained urban environment.