CV based
cyclist recognition

Abto Software was approached by a solution provider delivering innovations to businesses and governments. The company’s main goal is covering regional challenges by providing infrastructure services, business tools, Smart City and IoT solutions.

Our team has joined to assist the organization in designing a bicycle tracking system and improve cyclist safety. The cooperation has resulted in the successful development of a demo solution, which allows to monitor bikers, non-bikers, speed, and even helmets.

The project’s main goal was designing a solution for smart cyclist detection by leveraging artificial intelligence. To catalyze technological transformation, the company was pursuing the mission to popularize bicycle culture and construct an efficient cycling infrastructure.

Our team designed an AI based cyclist detection program that:

• Tracks bikers, non-bikers, speed and cycling helmets
• Ensures high recognition and calculation accuracy

We developed a solution, which enables interested parties to collect, store, access, and analyze traffic data. This data includes information about every single cyclist and non-cyclist that enters or exits the track.

This custom AI program can be used to send notifications about bikers not wearing their helmets or speeding. Such control can help drive interest in the cycling culture and improve public safety.

Here’s a step-by-step explanation of the solution’s operation:

1. The camera is placed directly above the track to monitor both bikers and non-bikers that enter and exit that track
2. The video is streamed from camera to workstation equipped with CPU/GPU hardware of sufficient processing power
3. The video is analyzed by the processing unit, which runs the application enhanced with AI algorithm
4. The implemented AI model recognizes bikers, differentiates bikers from non-bikers, captures whether recognized bikers wear helmets or not, and measures their velocity

During the solution’s prototyping, we covered:

a. Dataset preparation

At the first stage, to gather traffic information, a camera was placed directly above the multi-lane bicycle track. The camera was capturing both bikers and no-bikers, as well as identifying whether they’re wearing helmets and at which speed they’re moving.

b. Algorithm training in three successive stages:

• Biker and non-biker detection
• Helmet and haircut differentiation
• Speed recognition

During the course of the project, we have faced the following challenges:

• Accessing necessary data sets

Obtaining necessary data sets was troublesome because of some delays on the client’s side.

• Setting equipment

We provided detailed guidance on how to properly install the required equipment, as the company’s employees faced issues adjusting the correct angle and achieving high quality of the final video.

• Choosing the most suitable AI model

Some models we considered are better for accurate speed recognition and others for precise helmet detection. To ensure accurate calculation, our engineers gradually adapted and trained the selected AI model.

• Recognition nuances

The used AI model sometimes confused voluminous hairstyles with helmets, which affected overall precision. For greater recognition accuracy, our team has conducted additional work on training the algorithm.

The initiative has proven the feasibility of further software development of the AI supported tracking system. The client now has the opportunity to provide the product as a ready-made solution to leaders that specialize in improving public surveillance and safety.

The project provides for:

1. Higher competitiveness through selling a unique, ready-made solution
2. Public surveillance and safety through monitoring cyclist behavior
3. More interest in the cycling culture
4. Enhanced reputation and image of the client company by promoting healthy habits