Friday 21 March 2025
A team of researchers has developed a new approach to detecting and analyzing driving events, such as lane changes and overtakes, using dashcam footage. Their method involves creating a compact representation of the video footage, known as a motion profile, which can be processed quickly and efficiently.
The researchers used machine learning algorithms to analyze the motion profiles and detect specific patterns associated with different driving maneuvers. They found that by incorporating a special type of convolutional neural network, called CoordConv, they could improve the accuracy of their model even further.
The team tested their approach using a dataset of 416 dashcam videos, which were manually labeled with various driving events. The results showed that their method outperformed other state-of-the-art approaches in detecting lane changes and overtakes, with an average precision of around 40%.
One of the key advantages of this new approach is its ability to process data quickly and efficiently. This makes it well-suited for use in edge computing scenarios, where processing power may be limited.
The researchers hope that their work will contribute to the development of more advanced driver monitoring systems, which can help improve road safety by detecting potential hazards earlier. They also plan to expand their approach to detect other types of driving events, such as cut-ins and stop sign violations.
The team’s method involves creating a motion profile from each dashcam video, which is then analyzed using machine learning algorithms. The motion profiles are created by extracting key features from the video footage, such as the position and movement of vehicles on the road.
To improve the accuracy of their model, the researchers incorporated CoordConv, a type of convolutional neural network that takes into account the spatial relationships between different objects in an image. This allows the network to better understand the context of the driving events it is detecting.
The team tested their approach using a dataset of 416 dashcam videos, which were manually labeled with various driving events. The results showed that their method outperformed other state-of-the-art approaches in detecting lane changes and overtakes, with an average precision of around 40%.
The researchers also found that their approach was able to detect certain types of driving events more accurately than others. For example, they were able to detect lane changes with high accuracy, but struggled to detect overtakes on the left side of the road.
Despite these challenges, the team is optimistic about the potential of their approach to improve driver monitoring systems.
Cite this article: “Detecting Driving Events from Dashcam Footage Using Motion Profiles and Machine Learning”, The Science Archive, 2025.
Dashcam, Motion Profile, Machine Learning, Lane Changes, Overtakes, Coordconv, Convolutional Neural Network, Edge Computing, Driver Monitoring, Road Safety
