OVERVIEW

E-Scooter Parking Management System Using AI

The client, a micro-mobility company offering electric scooters, faced challenges with scooter parking, including unauthorized parking and inconsistent regulations across cities. Existing solutions like lock-based systems and navigation tools had limitations. To address these issues, we combined computer vision and machine learning to develop a solution. This solution included parking area detection using markers, proximity-based verification, and adaptability to diverse environments. Our comprehensive approach aimed to improve scooter parking management effectively.

Problem

The client faced significant parking challenges. Users frequently left scooters in unauthorized areas, resulting in fines and penalties from local governments. Initially, the client relied on lock-based systems, similar to those used by Lime, where users were required to lock the scooters at designated spots. However, many users ignored or misused these locks, parking scooters improperly and causing regulatory issues. Additionally, companies like Bird partnered with Google to develop navigation and VR-based solutions to guide users to correct parking spots and required users to upload photos for verification. Despite some success, these methods had limitations. VR solutions couldn’t pinpoint all locations accurately and struggled with the varied infrastructure and regulations of different cities. Furthermore, the client had limited data on parking environments and user behavior, which hindered the development of a comprehensive and effective solution.

Solution

To address these parking challenges, we developed an integrated solution combining elements from existing methods and introducing advanced technologies like computer vision and machine learning. Key features of our solution include:

 

  1. Parking Area Detection: Using signboards or specific markers, our system detects designated parking areas and verifies if the scooter is parked within these areas accurately through computer vision.
  2. Proximity-Based Verification: An algorithm identifies the closest scooter to the parking marker, capturing and analyzing images to ensure proper parking.
  3. Enhanced Lock-Based System: Retaining the lock-based system, we added an automated verification process. Users still lock the scooter, but the system captures an image and uses computer vision to verify its location.

  4. Improved Navigation Tools: Navigation tools were enhanced to provide real-time feedback and augmented reality (AR) cues, making it easier for users to find designated parking areas.

  5. Adaptability: The system adapts to different city regulations and parking infrastructures, using ground markers or other visual cues to determine correct parking in areas without specific stands.

     

    This comprehensive approach improved parking compliance, reduced fines, and provided a more reliable and adaptable solution for managing scooter parking.

Features & Benefits

Accurate Parking Detection

Ensures scooters are parked in designated areas using advanced computer vision.

Proximity-Based Verification

Accurately identifies and verifies the nearest scooter to the parking marker.

Adaptability

Can be customized to fit various city regulations and infrastructure setups.

Reduced Fines

Significantly reduces fines by ensuring compliance with local parking regulations.

Data-Driven Decisions

Utilizes data analytics to improve understanding of parking behaviors and environments.

Technologies Used to Build the Solution

Python, PyTorch, TensorFlow, OpenCV, YOLO, FastAPI, AWS (EC2 instances with GPU support), AWS for scalable deployment and handling computational loads