Our machine learning team consulted an IT company on implementing an AI solution counting road signs on the video to ensure safe highways.

Exposit specialists analyzed the client’s workflow, inspected the dataset, and provided detailed recommendations on overcoming pitfalls to reach project success.

Customer goal

Sometimes, implementing a custom software project requires engaging outside expertise in  addition to the knowledge of in-house specialists. It allows for increasing the solution’s effectiveness, optimizing the work process, and accelerating the desired result.

That’s how our client, an IT company providing digital products and services, addressed Exposit. The company requested our consulting service on a machine learning project in the highways and roads industry.

ML video analysis

The software analyzes a video from a video surveillance device with an ML model and counts streetlights, road signs, and crash barrier length to ensure safe roads.

Our job was to review the existing code and dataset, analyze workflow and techniques, and provide a client’s team with detailed recommendations on improving the solution implementation the fastest way.

Solution

Our work on the project was carried out in close and smooth collaboration with the client.

Firstly, we have received a document with a description of the software implementation. Our ML engineers conducted a thorough audit of the existing solution, a code, and a dataset and arranged a meeting with the customer to immerse in the project even more.

As a result of the teamwork, we have provided a document with the growth points of the project and detailed instructions for their implementation:

• Recommendations on the dataset review and relabeling. For example, one of the main pitfalls lay in the model training on pictures where the road sign took the whole area instead of using a realistic, complete image of sites;

• Guide with several approaches to the solution development with the pros and cons of each of them.

The offered methods allowed to use the existing code, as well as look at the new technologies ensuring better solution reliability and flexibility.

ML project steps

Our client was delighted with the project’s results and left their review on Clutch:

“Exposit’s specialists welcome a challenge, and they don’t look for easy solutions.

While it is customary for consultants to completely dismiss the work you have already done and propose a solution of their own, the Exposit team took the more difficult approach of taking what we had and making it better. They succeeded admirably.

In addition to fulfilling their duties, they also shared lots of useful advice on how to improve the quality of our work.”