Table of Contents

Applying AI and Machine Learning for Robust Transport Infrastructure

Robust transport infrastructure for goods and population is seen as a key driver for macroeconomic growth. Digital technologies like computer vision-based AI and machine learning can help monitor the quality and structural integrity of a vastly distributed network of roads, railway lines, bridges, transit stations, and cargo hubs. That helps in planning maintenance actions to ensure 100% availability of assets on critical network paths at optimal operational costs.

Roadways and railways are the primary modes of land-based transportation of goods and people around the world. Apart from the vehicles that ply these routes, the routes are key assets for the transport network. Due to extended and capacity usage, roads and railway lines suffer from surface quality issues and structural integrity issues.

Bridges, tunnels, and roadside slopes on highways in mountainous regions often suffer from structural integrity issues. The same is the case with conditions of partial or total impairment of these assets resulting in traffic congestion and delays, accidents, and route blocks. It reduces the overall transportation system’s throughput and subsequently reduces industry activity. This has huge macroeconomic implications and impacts the quality of life.

Understanding surface quality and structural integrity

Surface quality –

Usage-based wear and tear of roads and railway lines leads to depleted surface quality. This is seen in asphalt-based tar roads as well as concrete-based roads. Another reason for degraded road quality is road excavation work by municipal agencies along with other reasons such as extreme weather conditions. All these contribute to deformities in the road surface. Types of road deformities/damage include:

  1. Cracks (crocodile/ spider, longitudinal)
  2. Potholes
  3. Patches
  4. Damaged painted lane markings

Poor surface quality in terms of the large density of such road deformities causes vehicle damage, and excessive wear and tear on tires, suspension, and transmission systems. It is often the root cause of the loss of structural integrity of roads leading to accidents.

Structural integrity –

Assets across the road transportation network like roads, bridges, railway lines, and switches are subjected to varying load and weather conditions during their long operating life. While these are part of the periodic maintenance routines by the technical staff, there have been numerous instances of a breakdown of route segments like road patches, rails, connecting plates and bolts, spikes, bridge pillars and segments, and the surface of road-side slopes. Structural integrity issues can lead to accidents in vehicles, causing loss of cargo, and even human life.

Some of these assets are often very difficult to access for manual inspection and might need drones to inspect. Most of these are spread out over a large geographical area and thus need serious inspection effort to be spent regularly to ensure adequate coverage of all network assets.

How technology helps

Compute vision-based AI/ML pipelines are used on video data streams captured from stationary and mobile cameras, often stereoscopic i.e. depth sensing, to gauge the level of deformities in the granular segments of the transport network.

For example, such cameras can be mounted on the masts supporting overhead electric cables on the railway lines. Alignment and continuity (rail lines), and integrity (joints and fasteners) are some key physical attributes that such camera feeds can help monitor. Deep learning-based algorithms for semantic segmentation, denoising, anomaly detection, and object detection are commonly used for such analysis.

Similarly, cameras mounted on cars can help detect deformities like potholes, cracks, and damaged lane markings using deep learning-based AI and machine learning algorithms as mentioned above. Traditional ML models like edge detection and spectral segmentation can also be used. Upon capturing and processing the vision data, various forms of road damage are detected, and cloud-based applications can help assign the severity (high, medium, and low) of the deformities in the area in scope. Subsequently, entire roads in a region can be profiled for such damage.

Some practical considerations

Designing for diverse route types and regions

The damage varies by the type and configuration of roads/rail lines. Highways have heavy traffic w.r.t the number of vehicles and heavy commercial vehicles as compared to city roads. They also go through mountainous, hilly regions, as well as low-lying areas that have mandatory roadside slopes. That adds to the need for capturing additional video streams and detecting complex damage types.

Similarly, railway junctions, as well as rail routes in hilly/seaside regions have a higher probability of damage to rail, joins, and switches, due to exposure to adverse weather conditions like heavy rain, loose soil, and ambient humidity. A large network of cameras needs to be designed to effectively capture the damage to route elements. A network of edge gateways is also needed to orchestrate the video and telemetry data integration and analysis using AI and machine learning models.

Collecting and annotating diverse data

The data required for training vision-based AI and machine learning models is quite extensive. It spans across input modalities, ambient conditions, and field of view complexities. Low light, occlusion, overlap, and camera vibrations impact the quality of video data used for the training and optimization of AI/ML algorithms. That leads to inefficient and ineffective annotations. Covering all these scenarios in model training leads to unprecedented volume, velocity, and variety of training data to be processed. Data science and engineering teams often need to use workflow management tools for image data annotation and engage crowdsourcing models for tackling the throughput and scale requirements while creating the training data sets.

Wrap-up

Deep computer vision-powered AI and machine learning pipelines can significantly improve the accuracy and efficiency of condition monitoring of public transport infrastructure network line roads and railway lines. AI/ML algorithms for object detection and semantic segmentation are only a part of the solution, bigger factors impacting the performance are inference optimization, runtime configurations, and availability of comprehensive training data.

eInfochips has helped multiple innovative companies leverage AI and machine learning technologies from leading platform providers for building edge and cloud applications for worker productivity and user experience and subsequently maximized the value chain throughput.

Please reach out to us for AI and ML-based services for building your connected, smart product development initiatives.

Explore More

Talk to an Expert

Subscribe
to our Newsletter
Stay in the loop! Sign up for our newsletter & stay updated with the latest trends in technology and innovation.

Our Work

Innovate

Transform.

Scale

Partnerships

Device Partnerships
Digital Partnerships
Quality Partnerships
Silicon Partnerships

Company

Products & IPs

Services