DIY Self-Segregating Smart Bin

[abhirathkoushik]

Problem Statement: To build a smart bin that can be used practically to segregate waste in domestic and industrial waste segregation units.

Why this problem?
As a team, we wanted to find a solution to the common problems we experienced or witnessed in our society.

We noticed that the waste management system in Bangalore was unhygienic and inefficient. Segregation of waste has a lot of advantages. Biodegradable waste can be composted separately while metal and paper can be recycled. These benefits are lost when they are mixed up.

Hence, we decided to build a Self-Segregating Smart Bin, which can provide a practical way and feasible way of segregating waste.

Materials and Software Used:
Raspberry Pi 3B
Logitech Webcam
Servo Motor
SR04 Ultrasonic Sensor
Python
Tensorflow
OpenVino

Implementation:

1. Hardware:

We have used RaspberryPi as our Microprocessor, to control the sensors, webcam, servo motors, and to process the deep learning models.
2 Ultrasonic Sensors (HC-SR04) present at the platform level of the bin indicate the presence of waste on the platform. This then triggers the webcam to click a picture of the waste present on the platform.
This picture is then compared with the dataset and categorized into either Biodegradable or Non-biodegradable, by an application called TensorFlow, installed on the RaspberryPi.
Then, a servo motor (MG 995) accordingly tilts the platform to slide the waste item into that respective category.
This process repeated again, with control given back to the Ultrasonic Sensors, to identify the presence of the next item placed on the platform.

2. Software:

We have used Google TensorFlow, as the application to run our deep learning models.
We have used Inception V3 as our model, based on its Accuracy and Speed, after comparing this with other models and versions.
Inception V3 is now executed by TensorFlow, which is downloaded onto the RaspberryPi.
When a picture is taken by the webcam, TensorFlow runs the model to identify and categorize the image into either Biodegradable and Non-biodegradable
Based on this result, the servo motor is then programmed to tilt the platform.

In this way, we can design a smart bin that can segregate waste by itself into biodegradable and non-biodegradable.

[gautha91]

Very interesting. Thank you for sharing @abhirathkoushik . Do you have a model diagram or architecture diagram for the same?

[Adhithya_Viswanathan]

@abhirathkoushik amazing work , I have a few questions about the bin
1.How much will it cost to build the bin(is it affordable) and,
2. Will the bin be able to replace the common bins?

[abhirathkoushik]

Yes. Please refer to the picture below for the Flowchart of this project.

The 2 pictures below are of the prototype we built.

Thanks and Regards,
Abhirath Koushik

[abhirathkoushik]

1. The prototype we built was using a RaspberryPi as the main microprocessor. This costs Rs3000. The other expensive component was the webcam which costs Rs1200. Hence, the total cost comes to about Rs 6000.

But if we can find an alternative device that can handle machine learning models, or if we use a cloud service to do the processing, we can definitely reduce the cost. Similarly, we can experiment with cheaper webcams and test out the models.

If we can account for the microprocessor and webcam, we can reduce the overall cost of the smart bin.

2. I don’t think I can answer this question yet. We would have to test it out practically. It depends on factors like how fast and precise it can categorize the waste thrown.

Mixed waste is a huge factor that needs to be considered. If there is some food leftover or if 2 items are thrown simultaneously, the model may not work as described.

So once these limitations are accounted for, it might be possible for the smart bins to replace the common ones.

Thanks and Regards,
Abhirath Koushik