|
Getting your Trinity Audio player ready...
|
Singapore has emerged as a global leader in harnessing artificial intelligence (AI) to bolster safety measures across a wide array of industries and sectors. Through strategic initiatives and collaborations, the city-state has embraced AI technologies to proactively address safety concerns and enhance overall security standards.
In the expansive field of AI, the spotlight often shines on Large Language Models (LLMs) and powerful APIs. However, a pioneering project called Balefire diverges from this trajectory, delving into computer vision at the edge. Through a collaborative effort with the National Environment Agency (NEA) since its inception, Balefire has evolved into a sophisticated edge AI solution aimed at detecting smoking activities in public spaces.
Concluding its third iteration, Balefire is a milestone in the realm of edge AI. Its primary objective in this latest state is to assist NEA in identifying smokers in areas where smoking is prohibited. By analysing the frequency and patterns of smoking activities alongside overall foot traffic data, Balefire provides NEA with crucial insights to optimise the deployment of enforcement officers in these identified hotspots.
While the concept of smoking detection might seem straightforward, Balefire’s journey reveals a multitude of complexities. One of the primary challenges lies in the detection of cigarettes themselves – small, often partly occluded objects that can easily be mistaken for other everyday items. Additionally, existing cues like smoke detection or the glowing tip of a cigarette proved unreliable, necessitating a more nuanced approach.
Initially, it seemed logical to assume that pre-existing solutions for smoking detection would suffice. However, Balefire’s operational requirements demanded a level of versatility and real-time capability not met by off-the-shelf solutions. This realisation spurred the development of a bespoke solution tailored specifically to NEA’s unique operational demands.
The evolution of Balefire’s pipeline reflects a journey of iterative refinement. Beginning with head detection and heuristic-based filtering, the pipeline then incorporates object tracking and smoke classification to minimise false positives and double counting. This comprehensive approach ensures that alerts sent to NEA are both accurate and essential for enforcement operations.
A critical aspect of Balefire’s development is the consideration of operational needs. Traditional metrics like accuracy, precision, and recall are supplemented by novel metrics like strict precision, which accounts for the occurrence of double counting. These metrics serve as benchmarks for evaluating Balefire’s performance and guiding further refinements.
Balefire’s evolution has been accompanied by invaluable lessons. From the importance of revisiting old solutions to the necessity of embracing pragmatic approaches, each lesson has contributed to Balefire’s continued enhancement and effectiveness in real-world deployment.
As Balefire continues to push the boundaries of edge AI, it underscores the significance of tailored solutions that seamlessly integrate technical expertise with operational requirements. With advancements in edge AI technologies, opportunities for impactful applications in public service delivery are abundant.
AI stands as a prominent force at the forefront of technology, permeating virtually every sector of modern society. With its transformative capabilities, AI can be used to unlock new opportunities for innovation and growth.
OpenGov Asia reported that the National University of Singapore (NUS) has established the NUS AI Institute to harness the potential of AI to address complex challenges and enhance societal well-being. Under the leadership of Professor Mohan Kankanhalli, NAII is committed to accelerating AI research and its practical applications, fostering collaboration, innovation, and meaningful societal impact across various domains.
Another example is the construction industry which plays a vital role in Singapore’s development, with large-scale projects such as public housing, ports, and business districts shaping the urban landscape. Despite its significance, the sector faces challenges regarding worker safety.
Video analytics platform leveraging AI to enhance safety standards on construction sites addresses crucial concerns and promotes safer working environments for workers and stakeholders alike.
