The Mildura Regional Council, in partnership with Minnovation Technologies, successfully deployed an advanced portable video analytics system during the Show and Shine Easter Event in Mildura. This event, which attracts over 25,000 visitors across two days, presents unique challenges for crowd management, safety, and operational efficiency. Leveraging the power of AI and deep learning, the collaboration sought to address these complexities by developing a portable and sustainable solution for real-time analytics of large, dynamic crowds.

Deep Learning and AI Models

A significant aspect of this solution was its use of deep learning for real-time video analytics. The system’s core AI models were meticulously trained to handle a range of scenarios common in crowded public events. The models employed convolutional neural networks (CNNs) to detect and classify pedestrians and vehicles. To cope with the density and dynamics of the large crowd, the AI models utilized NVIDIA DeepStream, a powerful software development kit (SDK) designed for real-time video analytics. DeepStream leverages the GPU’s power to process video streams efficiently, enabling the system to handle high-throughput environments with multiple camera feeds. With support for sophisticated deep learning models, DeepStream provided rapid object detection, classification, and even anomaly detection capabilities, which were crucial for monitoring large, dynamic crowds at the event. The platform’s ability to integrate with different AI frameworks made it adaptable and highly efficient for complex, real-time video analysis tasks.

To ensure the system could manage varying crowd densities, models were trained using diverse datasets, which included images of events with fluctuating population densities and a range of object sizes. This helped the AI distinguish individuals and smaller groups, track movements, and detect potential safety issues, even in areas with heavy crowding. Moreover, the system was capable of identifying abnormal activities, helping event managers respond to potential risks promptly.

A critical component of the system was the people counting model, designed to accurately count individuals even in tightly confined crowds. This model utilized advanced deep learning techniques to differentiate between overlapping individuals and minimize occlusion issues, which are common in dense environments. The people counting model relied on a combination of object detection and density estimation to ensure reliable results, even when the crowd was at its peak. Additionally, a re-identification tracking algorithm was employed to further increase accuracy. This algorithm allowed the system to recognize individuals across different camera frames, reducing the likelihood of double counting and improving the overall precision of the population estimates.

The Nvidia GPU processing unit played a pivotal role in facilitating the deep learning models’ capabilities. It enabled parallel processing of multiple camera feeds, allowing for seamless tracking of objects in real-time without sacrificing accuracy. This processing power was critical, especially during peak hours when the density of attendees was at its highest.

Adaptive Camera Technology

The cameras used in this video analytics setup were dual high-definition 4K units equipped with adaptive imaging technology. One of the key challenges in outdoor events like the Show and Shine Easter Event is managing variable lighting conditions. The cameras were designed to adjust to these changes automatically, using dynamic exposure control and high dynamic range (HDR) capabilities. This adaptability was essential for maintaining consistent image quality during both bright midday hours and shaded or evening conditions, ensuring the AI models had clear visuals for accurate analysis.

The HDR feature allowed the cameras to capture details in both bright and dark areas simultaneously, which was particularly useful during transition times like sunrise and sunset, when lighting conditions can change rapidly. This capability ensured the video analytics system could maintain high performance, irrespective of the time of day or sudden changes in weather that affected lighting.

Impact and Broader Applications

The portable video analytics system demonstrated its efficacy in managing the large crowd of over 25,000 attendees at the Show and Shine Easter Event. It provided event organizers with crucial insights into population density, movement patterns, and demographics, enabling informed decision-making to ensure public safety and enhance the visitor experience. The system’s real-time monitoring capability meant that any potential overcrowding or safety hazards could be addressed proactively, minimizing risks.

The successful implementation of this solution has attracted interest from other councils across Australia, considering similar deployments for their events. The case study illustrates the potential for AI-driven, portable video analytics to improve the safety and efficiency of large public gatherings. The alignment with the Victorian State Government Department of Jobs, Precincts, and Regions’ Regional Digital Plan further highlights how such innovations can support regional development, promoting safer and more enjoyable community events.

Portability and Ease of Use

Another notable feature of the video analytics system was its portability. Designed to be deployed by a single operator within minutes, the trailer-mounted unit was both solar-powered and self-sufficient, reducing the logistical burden typically associated with video surveillance systems. The system’s mobility allowed it to be positioned strategically across the event grounds, optimizing coverage and ensuring that critical areas were effectively monitored.

This ease of deployment makes the solution particularly appealing for regional councils, where resource constraints often make it difficult to manage large public events effectively. By minimizing setup time and resource allocation, the system demonstrated how modern technology could enhance operational efficiency while reducing costs.

Conclusion

The deployment of the portable video analytics system at the Show and Shine Easter Event in Mildura serves as a benchmark for the effective use of deep learning and AI in public event management. The collaboration between Mildura Regional Council and Minnovation Technologies underscores the power of innovative partnerships in tackling complex challenges associated with large gatherings. This project not only improved public safety and event experience but also set a precedent for the broader adoption of AI-driven solutions in event management across Australia.

References

• State Department of Jobs, Skills, Industry and Regions
• Mildura Show and Shine 2022