AUTONOMOUS AND AI-ASSISTED TRAFFIC MONITORING SYSTEMS FOR ACCIDENT-PRONE HIGHWAYS

Background

Road traffic crashes constitute one of leading causes of death and disability in Nigeria, WHO consistently ranking Nigeria among countries with highest estimated road traffic fatality rates globally. Substantial proportion of crash severity, particularly on high-speed federal highways carrying mixed light and heavy-vehicle traffic, attributable not solely to initiating crash event but to delays in incident detection and emergency medical response during which treatable injuries may become fatal. On many Nigerian highway corridors, incident detection continues to rely primarily on passerby reporting to FRSC or Nigeria Police, process frequently subject to substantial delay, particularly during night-time hours or sparsely trafficked sections.

Advances in AI, particularly computer-vision-based object detection, have created new possibilities for automating aspects of highway incident detection previously dependent entirely on human observation. Modern object-detection architectures including YOLO (You Only Look Once) family capable of processing video streams real time to detect and classify vehicles, and with appropriate training data to recognize anomalous conditions such as stopped or overturned vehicles indicative of crash, at level of computational cost increasingly feasible for deployment on relatively low-cost edge-computing hardware suited to highway roadside installation.

Uyo-Itu Federal Highway, 38.4 km corridor linking Akwa Ibom State capital to Itu and onward to Cross River State boundary, identified in FRSC sector command records as one of more accident-prone highway corridors within Akwa Ibom State, carrying substantial articulated and heavy-goods-vehicle traffic alongside intercity passenger and light-vehicle traffic over geometric alignment including several curved and undulating sections associated with elevated crash risk. This study investigates design and evaluates potential performance of autonomous AI-assisted traffic monitoring system for early incident detection along corridor informed by systematic analysis of historical crash-record data to prioritize deployment at corridor's most accident-prone locations.

Study forms part of broader evolving national and global recognition that road-safety improvement in developing-country contexts increasingly depends not solely on physical infrastructure investment such as road widening or junction redesign but also on operational and technological interventions capable of reducing human and institutional response-time bottlenecks that frequently determine whether crash results in fatality, serious injury, or manageable outcome. Nigeria's National Road Safety Strategy alongside comparable frameworks adopted by numerous other countries increasingly emphasized 4 Es of road safety namely Engineering, Education, Enforcement and Emergency response with emergency response time recognized as area where relatively low-cost technological intervention exemplified by AI-assisted monitoring system examined in this study can yield disproportionately large safety benefit relative to capital cost, particularly when deployment precisely targeted at data-identified high-risk locations rather than pursued as undifferentiated corridor-wide investment.

Intelligent transportation system project topics | External: MIT - Intelligent Transportation Systems, FHWA - ITS and Safety, WHO - Road Traffic Injuries

Statement of Problem

Despite well-documented severity of road-safety challenge along Uyo-Itu Federal Highway and comparable Nigerian corridors, incident detection along most sections of highway network remains dependent on manual passerby-initiated reporting resulting in emergency response delays widely believed though rarely quantified in local context to contribute to preventable crash-outcome severity.

While AI-based traffic-monitoring and incident-detection systems have been demonstrated in research and in some cases deployed operationally in more developed transportation contexts, technical feasibility, expected performance, and cost of such system have not previously been systematically assessed for specific Nigerian highway corridor using locally collected crash and imagery data.

This study addresses problem by developing and evaluating AI-assisted incident-detection system design specifically for Uyo-Itu Federal Highway informed by rigorous crash-data-driven black-spot identification and locally trained computer-vision model covering six classes car, bus/coach, truck/trailer, motorcycle, pedestrian, stopped/crashed vehicle.

Aim and Objectives

Aim: to design and evaluate autonomous AI-assisted traffic monitoring system for early incident detection along Uyo-Itu Federal Highway, Akwa Ibom State, with view to reducing incident-to-emergency-response detection time at corridor's most accident-prone locations.

·         Analyze five years historical crash-record data for Uyo-Itu Federal Highway to identify and rank crash black-spot locations;

·         Develop and train YOLOv8-based computer-vision model capable of detecting vehicles and recognizing stopped/crashed-vehicle incidents from highway camera footage;

·         Evaluate trained model's detection performance on held-out test dataset;

·         Design proposed system architecture for autonomous incident detection and alert generation suited to deployment at identified black-spot locations;

·         Estimate through discrete-event simulation potential reduction in incident-to-alert time achievable under proposed system relative to prevailing manual-reporting baseline; and

·         Estimate capital and operating cost of deploying proposed system at identified black-spot locations.

Research Questions

·         Where are principal crash black-spot locations along Uyo-Itu Federal Highway based on historical crash-record analysis?

·         What level of detection performance can YOLOv8-based computer-vision model achieve for vehicle and incident recognition on this corridor?

·         What system architecture appropriate for autonomous incident detection and alert generation at identified black-spot locations?

·         What reduction in incident-to-alert time achievable under proposed AI-assisted system relative to prevailing manual-reporting baseline?

·         What is estimated capital and operating cost of deploying proposed system?

Significance

Significant because it provides Federal Ministry of Works, FRSC and Akwa Ibom State Government with technically grounded cost-estimated proposal for AI-assisted incident detection specifically prioritized at most accident-prone locations of documented high-risk Nigerian highway corridor, offering more targeted and evidence-based investment case than generic undifferentiated corridor-wide proposal.

Study also contributes original locally trained and evaluated computer-vision model and annotated dataset specific to Nigerian highway traffic conditions important contribution given most published object-detection models for traffic applications trained and validated on imagery from substantially different traffic and infrastructure contexts.

For broader research community, demonstrates replicable crash-data-driven methodology for prioritizing AI-based ITS investment that other Nigerian highway authorities can apply to own accident-prone corridors. Also significant at level of engineering education and capacity-building: by demonstrating locally relevant technically rigorous AI-based road-safety research project can be undertaken using open-source tools YOLOv8, SimPy, Python's scientific computing stack and modest hardware requirements, provides template that can inform design of comparable final-year and postgraduate research projects in Nigerian civil and computer engineering departments seeking to build applied machine-learning research capacity within civil-engineering context increasingly important interdisciplinary skill set as ITS and AI-based infrastructure applications continue to expand globally and within Nigeria specifically.

Computer vision and AI project topics | Civil engineering road safety topics

Scope and Limitations

Limited to 38.4 km Uyo-Itu Federal Highway corridor within Akwa Ibom State. Crash-data analysis covers FRSC-recorded crashes for five-year period 2021-2025. Model development and evaluation based on annotated image dataset extracted from available highway CCTV and dashcam footage covering six object/incident classes: car, bus/coach, truck/trailer, motorcycle, pedestrian, stopped/crashed vehicle. Study addresses design and simulated performance evaluation of proposed system; does not extend to physical field deployment, installation or field trial given capital investment and institutional coordination required beyond undergraduate research, though physical deployment recommended as subsequent phase.

Subject to limitations: historical crash-record data from FRSC while most authoritative available source subject to known general limitation of underreporting characteristic of Nigerian crash-record systems particularly for minor-injury and property-damage-only crashes meaning black-spot identification based on recorded rather than total crash incidence. Annotated training dataset comprising 4850 image frames while substantial for undergraduate research remains modest relative to scale of datasets used in state-of-the-art commercial traffic-AI systems, and model performance in full field deployment may differ from held-out test-set performance due to variation in lighting, weather and camera-positioning conditions not fully represented. Incident-to-alert time reduction estimates derived from discrete-event simulation informed by model detection-latency measurements and literature-reported emergency-response benchmarks rather than field trial measurement, and should be interpreted as estimate of potential performance rather than guaranteed field outcome.

Operational Definitions

Intelligent Transportation System: Application of sensing, communication and computational technologies to transportation infrastructure to improve safety, efficiency and mobility.

Black-Spot: Defined road location exhibiting statistically elevated concentration of crashes relative to comparable sections, identified through crash-record analysis - here 4 clusters accounting 41.2% fatal and serious-injury crashes despite 18.6% length.

Object Detection: Computer-vision task involving identification and localization of specific object classes e.g., vehicles, pedestrians within image or video frame.

YOLO: Family of single-stage real-time object-detection neural network architectures capable of simultaneous classification and localization - YOLOv8 used achieving mAP@0.5 89.7%.

Mean Average Precision: Standard evaluation metric for object-detection models summarizing precision-recall performance across classes and thresholds - 89.7% overall, stopped/crashed precision 84.3% recall 81.6%.

Edge Computing: Computing paradigm where data processing video inference occurs on or near device generating data rather than remote data centre reducing latency and bandwidth.

Incident-to-Alert Time: Elapsed time between occurrence of traffic incident and generation of actionable alert to emergency responders - principal metric: 14.2 minutes manual vs 47 seconds automated 94.5% reduction via simulation.

Empirical Bayes Method: Statistical method for black-spot identification that combines observed crash frequency with model-predicted expected frequency to produce more reliable estimate of underlying crash risk than raw counts.

Conclusion

Analysis of five years FRSC data via kernel density and Empirical Bayes identified four statistically significant black-spot clusters accounting 41.2% of recorded fatal and serious-injury crashes despite representing only 18.6% of total 38.4 km corridor length, providing data-driven prioritization. YOLOv8 model trained on 4850 annotated frames covering six classes achieved mAP@0.5 89.7% and stopped/crashed-vehicle precision 84.3% recall 81.6% on held-out test set, demonstrating technical feasibility with locally collected Nigerian highway imagery.

Proposed architecture combining roadside camera units, edge-computing inference nodes, centralized alert dashboard evaluated through discrete-event simulation indicates potential reduction in mean incident-detection-to-alert time from 14.2 minutes manual/passerby-reporting to 47 seconds automated, reduction 94.5%. Cost estimation indicates capital cost approximately NGN 187 million for full corridor coverage 14 camera/edge units with annual operating cost NGN 22.4 million.

Conclusion: AI-assisted incident detection both technically feasible and likely to yield substantial emergency-response time improvements along identified black-spot sections of Uyo-Itu highway. Recommend phased deployment beginning with four identified black-spot clusters, integration with existing FRSC emergency response protocols, and periodic model retraining using locally collected data to maintain detection accuracy over time.

FAQs

What is autonomous AI-assisted traffic monitoring?

System combining roadside cameras, edge-computing YOLOv8 inference, and centralized alert dashboard to automatically detect stopped/crashed vehicles and generate emergency alerts without manual reporting.

Where are black-spots on Uyo-Itu highway?

Four statistically significant clusters identified via kernel density and Empirical Bayes from 2021-2025 FRSC data accounting 41.2% of fatal and serious-injury crashes despite only 18.6% of 38.4 km corridor length.

What detection performance did YOLOv8 achieve?

Mean Average Precision mAP@0.5 89.7% across six classes car, bus/coach, truck/trailer, motorcycle, pedestrian, stopped/crashed vehicle; stopped/crashed precision 84.3% recall 81.6% on held-out test set of 4850 annotated frames.

How much time reduction does AI system offer?

Discrete-event simulation indicates reduction in mean incident-detection-to-alert time from 14.2 minutes manual/passerby to 47 seconds automated - 94.5% reduction.

What is system architecture proposed?

Roadside camera units capturing video, edge-computing nodes running YOLOv8 inference locally to reduce latency and bandwidth, centralized alert dashboard integrated with FRSC emergency response protocols.

What is cost of deploying AI traffic monitoring on Uyo-Itu?

Capital cost approximately NGN 187 million for full corridor coverage 14 camera/edge units, estimated annual operating cost NGN 22.4 million - targeted deployment at four black-spots recommended for phased investment.

What is Empirical Bayes black-spot method?

Statistical method combining observed crash frequency with model-predicted expected frequency to produce more reliable estimate of underlying crash risk than raw counts alone, used with kernel density for hotspot ranking.

What are limitations of FRSC crash data?

Underreporting characteristic of Nigerian crash-record systems particularly for minor-injury and property-damage-only crashes, meaning black-spot identification based on recorded rather than total crash incidence.

Why use locally trained YOLOv8 model?

Most published traffic object-detection models trained on imagery from different traffic and infrastructure contexts; locally trained on Nigerian highway CCTV/dashcam captures local vehicle mix, lighting, road geometry for better accuracy.

What recommendations does study make?

Phased deployment beginning with four identified black-spot clusters, integration with existing FRSC protocols, periodic model retraining using locally collected data to maintain accuracy, subsequent field trial beyond simulation.