Problem

In a manufacturing line, a Computer Vision system is used to identify and sort defective products. The accuracy of this system was analyzed and it was found there was a high count of false positives. A solution was required to reduce the number of false positives.

Solution

We ran various clustering machine learning algorithms on the Computer Vision system prediction dataset and derived the maxima and minima.

This input was used to optimize the Computer Vision system algorithm by setting a threshold and thus reduce the number of false positives.

Problem

Today we can make security cameras, drones and other utility appliances intelligent by integrating AI algorithms on them. But running these algorithms are computationally expensive. So the edge devices that run these algorithms in these appliances are priced at a premium and out of range for anyone who just wants to experiment or do a Proof of Concept for a client.

Solution

The objective was to provide a cheaper solution to run AI algorithms in a camera device that could detect and identify people from a known group. For this we used a lightweight pre-trained model and trained it with images of people from the target group. This model was deployed in a basic edge device kit attached to a camera. The kit allowed us to control basic movements of the camera using PID controller. So we had a camera device that could adjust its field of vision as well as detect and identify a person.

Problem

Every year people and businesses lose hundreds of hours in transit. This results in lower productivity and revenue loss.
For this, we need to predict traffic congestion on routes used for transit.

Solution

First, we use an auto encoder (deep learning algorithm) to reduce the size of very large map data.
Then recognize congestion pattern changes in the reduced data and predict future congestion points from the current situation using time-series analysis.