End-to-end pedestrian collision warning system based on a convolutional neural network with semantic segmentation
DOI: 10.1109/icce.2018.8326129
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the high false alarm rate inherent in traditional pedestrian collision warning (PCW) systems, which often trigger warnings in safe scenarios, such as when pedestrians are walking on sidewalks. These false alarms distract drivers and reduce system reliability. To solve this, the authors propose an end-to-end PCW framework based on a convolutional neural network (CNN) that directly predicts danger from raw images, eliminating the need for a separate, error-prone pedestrian detection stage. The proposed architecture combines two networks: a prediction network for binary classification of dangerous situations and a semantic segmentation network that extracts contextual features. These networks share low-level convolutional layers to reduce parameters and are trained simultaneously by minimizing a combined loss function consisting of cross-entropy for prediction and Euclidean loss for segmentation. The authors address class imbalance in the training data by duplicating warning-case images to match the volume of non-warning cases. The model is trained on the Cityscapes dataset, using 2,975 densely annotated training images and 1,525 test images, with manual annotations for warning labels. Experimental results demonstrate that the proposed method significantly outperforms a traditional Histogram of Oriented Gradients (HoG)-based baseline. At a 15% false positive rate, the HoG-based system achieved 45% accuracy, whereas the proposed CNN without semantic segmentation reached 63.94%, and the full model with semantic segmentation achieved 71% accuracy. Receiver operating characteristic curves further confirm that the proposed method, particularly with semantic segmentation, offers superior performance across most false positive rates. Qualitative results indicate that the system effectively distinguishes between dangerous and safe scenarios for both pedestrians and cyclists, avoiding unnecessary alarms when subjects are on sidewalks. The study concludes that end-to-end CNN-based PCW systems are feasible and effective at reducing false alarms while improving warning accuracy. By leveraging semantic segmentation, the system implicitly classifies objects and understands scene context, leading to more reliable decisions. The authors suggest that this framework could be extended to other advanced driver assistance systems, such as lane departure and forward collision warnings, and note that future improvements could involve adopting more recent deep learning architectures like residual networks.
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | OpenAlex-citations | — | — | 1 | 2026-06-20 |
| archive | success | semantic_scholar | — | — | 6 | 2026-06-26 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-20 |
| chunk | success | chunk | — | — | 1 | 2026-06-20 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-20 |
| promote | success | — | — | — | 1 | 2026-06-20 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-20 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.