Automatic detection of road hazards for safety and early removal

Early detection of obstacles and dangerous objects present in the road can reduce the risks related with these hazards. It enables fast removal of such objects and allows all parts involved, motorway operators, drivers, police forces and first responders teams to receive detailed alerts so they know about what dangers lie in the road and can act accordingly.
Our proposal uses the latest developments in image processing and Artificial Intelligence to detect road hazards, assess the risk that each hazard presents and alert the different parts involved: motorway operators, drivers, police forces and first responders teams.
Our Solution uses a network of connected cameras for motorway monitoring. Already deployed IP cameras can be used and more can be installed in the existing infrastructure to extend the network coverage. Each camera position will be known so each detection can be precisely positioned along the road.
All the heavy image processing is hardware accelerated. All images produced by the cameras in the network are be processed in specialized hardware that facilitates and accelerates the processing demands of AI tasks. This hardware is centralized and can be on premises or in the cloud. This allows to use simple sensing hardware, cameras with no processing capabilities, and facilitates software maintenance, software update and scalability of the pipeline.
Optimized AI-based multi-sensor processing for image understanding. Our proposal uses a stream-processing pipeline that captures and processes images from the network of connected cameras. AI model inference is hardware accelerated so results can be obtained in real time.
Many different AI models can be run simultaneously. With this approach we are able to run different AI models that excel in different detections: models for main obstacle detection, for small litter detection or for the detection of plastic films or fabrics. Each detection image area can be further processed to infer other properties like flammability or toxicity. Objects can be tracked, so obstacle persistence is ensured. At the end of the stream-processing pipeline, any detected object will be fully characterized.
Last part of our solution consists of an alert system that collects all the detections information resulting from the AI models to communicate the necessary alerts to the differnt parts:
- Motorway operators can receive continous information of obstacles and potentially dangerous objects detectedin real time. They can then choose the best way to proceed in each case.
- Drivers can also be alerted of obstacles or hazards that affect the driving. Message boards can show information about what obstacles are present in the road and where. Motorway information radio stations and webs can also be updated with such information.
- Finally, police forces and first responders can be alerted about high risk obstacles, fires or conditions that might require their assistance.