Smart Garbage Visual Detection, Monitoring and Analytics – The MANGUSTA
Garbage and generic waste management is a challenging task in modern cities. Every area has its peculiar waste production pattern in terms of kind and volume of produced waste, and optimizing collection is key to reduce costs and ensure at the same time that city decor is always maintained.For some cities, this task is made even more difficult due to the impossibility of installing underground containers. This is the case of Amsterdam, where in most part of the city center, garbage collection relies on citizens and tourists to drop trash bags at given collection spots, at given hours (twice a week). In this case, it is of course vital to optimize the collection process and to minimize the amount of trash bags accumulating at any of these spots.Many projects that aim to solve this problem involve some form of sensors to be scattered through the city, which would be responsible to collect data about garbage distribution (IoT-style). We find this approach expensive, both for installation and maintenance, not at all scalable and not environmental-friendly. The solution to environmental problems cannot be to produce and scatter even more disposable electronics all over a city. –> More info on: https://becominghuman.ai/smart-garbage-visual-detection-monitoring-and-analytics-a0061fff2b76
YOLO – AI object detection
You only look once (YOLO) is a state-of-the-art, real-time object detection system. On a Pascal Titan X it processes images at 30 FPS and has a mAP of 57.9% on COCO test-dev.
How It Works
Prior detection systems repurpose classifiers or localizers to perform detection. They apply the model to an image at multiple locations and scales. High scoring regions of the image are considered detections.
We use a totally different approach. We apply a single neural network to the full image. This network divides the image into regions and predicts bounding boxes and probabilities for each region. These bounding boxes are weighted by the predicted probabilities.
https://pjreddie.com/darknet/yolo/
Other object detection AI with AR
1) ARkit: Record spatial features of real-world objects, then use the results to find those objects in the user’s environment and trigger AR content
In iOS 12, you can create such AR experiences by enabling object detection in ARKit: Your app provides reference objects, which encode three-dimensional spatial features of known real-world objects, and ARKit tells your app when and where it detects the corresponding real-world objects during an AR session.
https://developer.apple.com/documentation/arkit/scanning_and_detecting_3d_objects
2)Spark AR: uses target market like vuforia
3) Vuforia: object recognition
Object Recognition allows you to detect and track intricate 3D objects, in particular toys (such as action figures and vehicles) and other smaller consumer products. Use the Object Scanner and the accompanying object target scanning image to easily scan your detailed toys, models, and educational tools.
Object Targets should be viewed indoors under moderately bright and diffuse lighting. To the extent possible, the surfaces of the object should be evenly lit and not contain shadows caused by other objects or people. This should also be accounted for when scanning the object.
For Object Recognition to work well, the physical object should be:
- Opaque, rigid and contain none or only very few moving parts.
- The surface of the object should have a large number of contrast-based features and rich texture.
https://library.vuforia.com/features/objects/object-reco.html
Conclusion:
Ar and object detection requires AI and an algorithm to be trained by millions of trash images (machine learning + python). Therefore it would be best to stick to 3D design of different types of trash and adding them to the experience like we did last semester