Understanding distracted pedestrian behaviors is critical to road user safety and preventing traffic-related injuries. Front-facing vehicle cameras (a.k.a., dashcams) have increasingly become popular for documenting driving behavior and patterns. However, a relatively underexplored application of dashcam footage is to automatically identify distracted road users that may pose a threat to pedestrians. Detecting distracted behaviors in dashcam-captured imagery can enable drivers to take preventative measures and avoid potential traffic accidents. To this end, computer vision techniques powered by the prediction capability of artificial intelligence (AI) can be leveraged to identify pedestrians’ distracted behaviors when crossing streets and intersections. In this paper, pedestrians’ phone-related distracted behaviors are detected and classified in dashcam footage by leveraging convolutional neural networks (CNNs) assembled in the form of a two-stage detection and classification architecture. In particular, we propose to first detect pedestrians (stage 1) followed by classifying the most prevalent distracted behavior visible in each detected instance (stage 2). This technique has been developed on an in-house video dataset collected from urban intersections around a major university campus. Results indicate that the pedestrian detection model achieves 76% average precision (AP), and the classification of distracted behavior reaches 72% precision, 98% recall, and 83% F1-score.
Pedestrian Phone-Related Distracted Behavior Classification in Front-Facing Vehicle Cameras for Road User Safety
ASCE International Conference on Computing in Civil Engineering 2023 ; 2023 ; Corvallis, Oregon
Computing in Civil Engineering 2023 ; 418-425
2024-01-25
Conference paper
Electronic Resource
English
DOAJ | 2021
|Wiley | 2021
|Mobile telephones, distracted attention, and pedestrian safety
Online Contents | 2008
|Mobile telephones, distracted attention, and pedestrian safety
Online Contents | 2008
|