Do you ever wonder what happens to the litter on the streets after you throw it away? In most cities, street cleaning requires a lot of resources and effort. However, Cortexia
, a world-leading solution, is determined to tackle this problem efficiently and sustainably with its computer vision system. Recently, students from Constructor Learning helped develop the second stage of the product called "dark zone analytics" (you can read the full blog post here
Cortexia has achieved remarkable results with its technology, and in this blog post, we will dive into how it works and its impact on city cleanliness.
Cities around the world face many challenges, one of them being the maintenance of cleanliness. Cortexia offers a computer vision system mounted on street sweepers and other vehicles to detect and count various types of litter left on the streets. The system utilizes data from deep learning inference taken daily in different city regions and applies machine-learning algorithms to predict the number of litter in areas uncovered by cameras.
The development of the product
The development of the Cortexia product was broken down into three stages: monitoring, prediction, and management
Monitoring refers to observing and tracking litter data and its aggregation as the Clean City Index (CCI). The CCI is scaled on human perception to ensure that the cleanliness as perceived by the citizens remains homogeneous within the targeted range. In the context of the product development mentioned, monitoring refers to the use of cameras and object detection to map the cleanliness level of a city.
Special cameras were installed on various modes of transportation like commercial vehicles, street sweepers, public transport, and bicycles. The aim was to create an accurate representation of the city's cleanliness, which was referred to as the Clean City Index. The data collected from the cameras would then be used to identify areas in need of improvement and to monitor the effectiveness of cleaning efforts over time.
To accurately predict the amount of litter in the city, it was important to account for the "Dark zones" that were not covered by the limited number of moving cameras, as the coverage of moving cameras in space and time is limited to around 30-40%.
To accomplish this, we enlisted the help of Constructor Learning students to develop a machine learning algorithm that could sample data from one part of the city and predict the amount of litter in other uncovered areas. To improve the algorithm’s accuracy, they integrated data from various measurement systems and additional features such as weather conditions and proximity to amenities like restaurants or bars. After training and evaluating multiple machine learning models, a data science pipeline was developed for predicting the amount of litter on the streets.
The noise in the data made it challenging to achieve high accuracy, but the model performed better when applied to larger street segments instead of just individual ones. By using this method, we are better equipped to determine the optimal street sweeper route for cleaning litter in the city, reducing the required resources and ultimately making the cleaning process more efficient.
Notable successes after the implementation of some tools:
Dynamic Cartography of Litter
Dynamic Cartography of Litter is a technology developed by Cortexia that provides real-time mapping of litter in a city and helps fill the streets’ gaps that traditional camera systems cannot reach. By implementing this technology, Cortexia has achieved notable success in litter management. The dynamic cartography system provides a comprehensive view of litter distribution in the city, making it easier for city authorities to monitor and manage the problem effectively.
The prediction tool uses machine learning algorithms to analyze images or data inputs of litter and categorize them into different types of waste, such as plastic, paper, glass, etc. This tool can be useful in helping individuals, organizations, and governments keep track of waste and manage it more efficiently. The demand for such a tool may come from the need to reduce litter and improve waste management practices.
Accuracy in forecasting
Accuracy in forecasting refers to the ability to predict future outcomes with high certainty. In the context of litter volume distribution, accuracy in forecasting can help a company like Cortexia optimize the routing of its sweepers. With accurate forecasts, Cortexia can plan the deployment of its sweepers more effectively and efficiently, leading to cost savings and improved overall performance. This external data serves as a complement to the spatiotemporal extrapolation performed by the system.
The effectiveness of the Dynamic Cartography of Litter system, a machine learning model used for prediction, is enhanced through the use of external data. Along with the spatiotemporal analysis, the system incorporates data on amenities and weather conditions to provide a more accurate and comprehensive picture of litter distribution. This integration of external data results in a more efficient and targeted cleaning process, as authorities are equipped with precise information on where and when litter is likely to accumulate. As a result, the system reduces the number of resources required for cleaning, saving time and money while helping to keep the city cleaner and more attractive for residents and visitors alike.
Management refers to the administration and control of an organization or system. In the context of product development, management refers to the processes and procedures put in place to ensure the efficient and effective operation of the product.
The management stage involves creating a management system that will optimize the routing of sweepers in the city cleaning process. Cortexia plans to develop an ERP (Enterprise Resource Planning) system to manage sweepers’ deployment, aiming to maximize efficiency and effectiveness in cleaning the city. The ERP system will help ensure the available fleet of vehicles and operational teams are deployed to the right places at the right time and the fleet’s maintenance is performed during off-peak hours, resulting in optimal cleaning results and efficient allocation of resources.
Cortexia's innovative solution is an excellent example of how technology can make our cities cleaner and more sustainable. With the help of Constructor Learning students, they were able to develop a machine learning model to predict litter in "Dark zones." By incorporating external data, like amenities and weather conditions, Cortexia's prediction stage becomes more accurate and efficient. The implementation of their Dynamic Cartography of Litter technology provides a comprehensive view of litter distribution in the city, making it easier for city authorities to monitor and manage the problem effectively. Cortexia is leading the way in clean city solutions, and we look forward to seeing how they continue to develop and innovate.
We at Constructor Learning are particularly proud of what Dominik
achieved with their capstone project. They significantly contributed to Cortexia’s efforts to tackle the city’s street litter problem. The machine learning model they created to predict the amount of litter in the “Dark zones’’ is a remarkable achievement and has the potential to improve the efficiency and effectiveness of city cleaning significantly. The students have demonstrated their expertise in data science and have shown that they have what it takes to make a real impact in the field.
The project was made possible by Cortexia, and we extend our special thanks to Julien Dupont
for his guidance. We look forward to working together in the future.