Visual Inspection

MODELS & INTEGRATION

Gauge Reading Code:

CLASSES =
[“person”, “hard hat”, “vest”]
image = cv2.imread(args[“image”])
detections = net.forward ()
label = “// “,CLASSES:”, confidence *
100,”% confidence”)
[“person”, “hard hat”, “vest”]
image = cv2.imread(args[“image”])
detections = net.forward ()
label = “// “,CLASSES:”, confidence *
100,”% confidence”)
[“person”, “hard hat”, “vest”]

Gauges card image

Reading
Clasifications: Safe

Gauge Type
Linear
Reading
29.85mmgh

Maintenance Repairs:

Scheduled
68
Completed
52
Electric power plant card image

Mission Reading:

Antenna
Unit 22
Reading
No Anomaly

Site Notifications:

Operational
34
Safety
2
Electric thermal card image

Reading
Clasifications : Safe

Safety Threshold
74.0 — 98.0
Reading
86.4

Automated Missions:

Operational
34
Safety
2

Mission Reading:

Antenna
Unit 22
Reading
No Anomaly

CORE USE CASES:

Whether it’s equipment at a facility, a storm-wrecked area or across vast acres of weather-impacted land, Levatas builds ROI-driven programs for all kinds of inspections.

Our models work with a wide variety of visual capture platforms, ranging from CCTV and single mounts, to advanced quadruped robots and drones, from traditional RGB cameras to hyper-spectral imaging systems including thermal and X-ray. With the right amount of data from cameras that provides quality images, our models bring the intelligence to see “beyond the pixels” and draw rich insights from the data.

Gauge iconAnalog Gauge Reading
SPOT Gauge image

With custom payloads on advanced robotics or drones, such as Boston Dynamics’ SPOT or DJI’s Inspire 2, customers are able to autonomously access more gauges in hard to reach places.

Levatas’ gauge reading model has the intelligence to adapt to and work in a broad range of real world conditions and environments, gauge types and overcome common technical challenges like glare, condensation and perspective warping. takes environmental challenges into account, recognizing that one reading may not be as straightforward as the next. When presented with challenges like an obstructed view, condensation or glare where a clear reading cannot be made, the model will proactively seek human-in-the loop feedback from an expert over low accuracy results.

Thermal Anomaly iconThermal Anomaly Detection
Thermal Anomaly detection image

Thermal Anomaly Detection can be deployed and optimized for any number of industries. Using thermal sensor information, analytics, and machine learning to identify specific objects and read their temperature against expected conditions, we help our partners find anomalies before they become problems.

The specific application of Thermal Anomaly Detection starts with your business’ challenges. Thermal cameras and robot payloads paired with our Intelligence Models can be used in cases like performing mass screenings of elevated body temperatures in high traffic places or to monitor equipment to ensure is operating within safe operating temperatures. Our models are able to blend thermal anomaly detection with classical object detection to discriminate even the hard cases and pinpoint zones within an image where a specific piece of equipment needs to be monitored, and ignore other noise in the picture which should be considered normal.

Unsafe Condition DetectionUnsafe Condition Detection
Unsafe Condition Detection

Computer vision is making the world safer. Whether it’s detection of dangerous situations like standing water or unstable structures, or calling attention to risky patterns of behavior, AI can alert teams to threats and reduce risk quickly and effectively.

For example, prove and maintain your compliance and keep your employees safer with our custom computer vision models that can detect PPE violations. Using state of the art computer vision techniques, our models are able to learn a baseline model of what normal operating conditions look like and can proactively alert humans when potentially anomalous conditions are encountered – even/especially if those conditions have not been encountered in the past. This gives our models maximum flexibility to adapt to real world environments where it may not be possible to pre-train models for every type of anomaly they may encounter.

Prediction Equipment iconPredictive Equipment Maintenance
Prediction Equipment Maintenance

Powered by our Vinsa Intelligence Models, our AI can detect when equipment at a facility shows signs of stress, overuse or anomalous behavior. Primarily used in manufacturing and assembly, the CV models we’ve deployed, can capture data using mobile or fixed cameras and flag issues to site supervisors, preventing costly line shutdowns or personal injury.

Using ensemble techniques, our models are able to blend structured and unstructured performance data together to identify equipment faults or anomalies allowing for proactive rather than reactive maintenance, improving plant safety and efficiency and extending equipment lifetime.

Prediction Equipment iconAerial & Satellite Imaging Analysis
Aerial and Satellite Image

Already collecting and analyzing images captured by plane flight, drone or satellite? Completely streamline the process by running CV models that, when trained what to detect, can take a fraction of the time. Like finding a needle in a haystack, technologies use our human-in-the-loop approach that provides analysts areas of focus, making the haystack much smaller, getting to results more quickly and offering a chance to apply human skill to more important aspects of their jobs.