Answers are grounded in this paper and cite their sources.
The growing incidence of forward collisions due to driver inattention highlights the critical importance of Advanced Driver Assistance Systems (ADAS) in enhancing vehicle safety through a robust Forward Collision Warning System (FCWS). Conventional techniques often rely on dual cameras or a combination of LiDAR, radar, and camera systems, increasing hardware cost and complexity. In contrast, this design uses a single camera for vehicle detection, tracking, and Time-to-Collision (TTC) calculation. A detailed comparative analysis revealed that the custom-trained You Only Look Once (YOLO) vehicle model optimized with TensorRT outperformed MobileNet Single Shot Multibox Detector (SSD), Region-based CNNs (R-CNN), ResNet-50, and ResNet-101, excelling in inference time, mean Average Precision (mAP), resolution, and initiation time. The model achieved an mAP of 65.8% with an inference time of 22 ms on the Berkeley DeepDrive (BDD 100K) dataset. Centroid-based tracking is employed to monitor detected vehicle centroids across consecutive real-time camera frames. The vehicle’s absolute speed, required for TTC calculation, is obtained via the OBD-II interface, enabling precise determination of the speed of the camera-equipped vehicle for accurate collision warnings. The results demonstrate that the low-cost single-camera system, enhanced by advanced optimizations and NVIDIA Jetson TX2 hardware, provides both robustness and reliability for FCWS applications.
No takes yet. Share an insight, caveat, or question.
Nosher et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: