# Detection of Micromobility Vehicles in Urban Traffic Videos

**Authors:** Khalil Sabri, Célia Djilali, Guillaume-Alexandre Bilodeau, Nicolas Saunier, Wassim Bouachir  
**arXiv:** [2402.18503](https://arxiv.org/abs/2402.18503)  
**Date:** 2024-02-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2402.18503v2/

## Key points

- Detection of objects in urban traffic environments is a significant challenge, especially with the increasing popularity of micromobility vehicles like e-scooters and bikes.
- Researchers from Polytechnique Montréal and the University of Québec (TÉLUQ) have developed an innovative detection model that integrates precision, speed, and enhanced features by applying aggregated feature maps from consecutive frames processed through motion flow to the YOLOX architecture.
- The adapted model introduces a temporal perspective to YOLOX's detection capabilities, leading to a deeper understanding of urban mobility patterns and significantly enhancing detection reliability.
- Testing on a custom dataset for urban micromobility scenarios showed impressive results, with substantial improvements over existing methods, highlighting the importance of considering spatio-temporal information for detecting small objects like micromobility vehicles in complex urban settings.
- The approach excels in challenging conditions such as occlusions and motion blur while ensuring temporal consistency, crucial for enhancing safety and efficiency in urban transportation systems where accurate object detection is essential.
- This research represents a significant advancement in improving object detection capabilities in urban traffic environments, particularly for micromobility vehicles. The integration of temporal information into traditional frameworks offers new possibilities for enhancing system performance and reliability in dynamic urban settings.

## Summary

In the realm of urban traffic environments, the detection of objects poses a significant challenge. With the rise in popularity of micromobility vehicles such as e-scooters and bikes, this challenge becomes even more pressing. To address this issue head-on, a team of researchers from Polytechnique Montréal and the University of Québec (TÉLUQ) have developed an innovative detection model that seamlessly integrates precision and speed with enhanced features. The core innovation lies in applying aggregated feature maps from consecutive frames processed through motion flow to the YOLOX architecture. By fusing these elements together, a temporal perspective is introduced to YOLOX's detection capabilities. This not only enables a deeper understanding of urban mobility patterns but also significantly enhances detection reliability. To test the efficacy of this adapted model, a custom dataset specifically curated for urban micromobility scenarios was used. The results were impressive, showcasing substantial improvements over existing state-of-the-art methods. This underscores the importance of considering spatio-temporal information when detecting small and thin objects like micromobility vehicles in complex urban settings. Furthermore, this approach excels in challenging conditions such as occlusions and effectively mitigates motion blur while ensuring temporal consistency. These advancements are crucial for enhancing safety and efficiency in urban transportation systems where accurate object detection is paramount. Overall, this research represents a significant step forward in improving object detection capabilities in urban traffic environments, particularly with regards to micromobility vehicles. The integration of temporal information into traditional object detection frameworks opens up new possibilities for enhancing overall system performance and reliability in dynamic urban settings.