Success Story: Transforming Safety Operations for a Waste Processing Company with NexTurn Vision AI

NexTurn has achieved a remarkable milestone by developing an automation solution that detects safety violations in a client’s warehouse facility. Leveraging the power of Meraki Cameras and advanced AI/ML models, our team successfully created a system that not only identifies safety breaches but also alerts stakeholders with timestamped evidence. This proof of concept demonstrates the potential of deploying cutting-edge technologies in client operations, transportation, logistics, and environmental health and safety (EHS) initiatives.

The primary objective of the project was to test the feasibility of deploying AI/ML models on Meraki Cameras. The client had already installed Cisco Meraki Cameras at various locations capable of running compressed AI models (TensorFlow Lite files) to generate inferences from live feeds. To achieve this, our team installed an MQTT broker to read messages and generated inferences from the camera and set up an AWS server instance to host the MQTT broker, maintain logs, and execute code.

The journey wasn’t without its obstacles. The team read through the different research papers and analysed various efficient AI Object detection models that can be modified to support camera-compatible files through previous practical experience. They carefully sorted through the AI object detection models by experimenting with the images collected/gathered and tracking model performance metrics for each iteration. Our team needed to convert this model to Tflite, figure out screenshot extraction with the timestamp (API call returned a null value for the timestamps initially), and integrate Python scripts with Meraki outputs.

After facing initial hiccup with the latest Tflite conversion, team decided to create a new Python virtual environment with downgraded libraries for Tflite conversion. Few more roadblocks such as time zone difference between where Meraki camera is located vs when Meraki API is used to fetch the screenshots, and limitations of the out of the box Meraki APIs, and Dashboard for the client use case were solved with bespoke engineering. Team breathed and lived the Agile way adopting Fail Early, Learn Early which helped our team to continuously deliver incremental value to our client.

Despite the challenges, our team completed the project within a remarkable 4-week timeframe! Our team’s proficiency in building deep learning models, intelligent dataset annotation, and meticulous documentation of test scenarios played a pivotal role in the success of this proof of concept. The successful execution of this project has opened doors for a full-fledged project with NexTurn. Stay tuned for more success stories as we continue to push the boundaries of innovation in niche technologies!