Our client is a leading provider of auto-load garbage trucks. With installed video cameras on their body, these trucks pick up the garbage bins from commercial places. Shops avail different sizes of bins as per their requirement. All trucks have sensors built in their chassis.
When pick-up happens, the video camera records the process, and someone has to review the video to confirm the pick-up manually.
Overflowing bins are subjected to penalty, whereas invoicing requires the exact timing of pick-up. The whole process was performed manually, and the client wanted to automate this manual review process of the garbage truck pick-up.
- Irregular transmission of data from sensors
- Classification algorithm on time-series data
- Extremely imbalanced sample (0.3% Target Variable)
- Analyzed 1.3 GB of input data covering 1.4 M records and 85 different signals from the sensor data
- Developed a model to analyze the images uploaded by the user and predict if the bin is overfilled or not
- Predicted the pick-up time within 5 seconds of vicinity based on sensor data collected from the truck’s chassis. Using Shiny Apps, the users were able to view the path taken by the truck, location of the overfilled bins as well as identify defaulters whose bins have been consistently overfilled
- The workload of the reviewer has been reduced to a few minutes
- Reviewer now has to watch at most 1.5 hours of clips/images of video per truck compared to reviewing 10 hours of the video earlier