Artificial intelligence
Smart Passenger Counting with an Overhead Camera
On-board computer vision that counts who boards through the rear door and keeps a video clip of every entry
- Client
- Urban transit operator
- Location
- Cartagena, Colombia
- Duration
- Since 2026
- Status
- Pilot on the road
Project Gallery
Technology Stack
- Python
- YOLO11
- Raspberry Pi 5
- Hailo-8L
- ESP32-S3
- AWS IoT Core
- AWS Lambda
- DynamoDB
- Amazon Cognito
- Kinesis Video Streams
- Terraform
- Next.js
Detailed Description
We designed and built a passenger counting system for city buses. A ceiling-mounted camera right above the rear door looks straight down while an on-board computer detects every person who crosses. The system counts entries, records a short clip of each one and sends it to the cloud, where the operator reviews events from a web dashboard. All counting happens inside the bus: it keeps working without signal and syncs when the connection returns.
The Challenge
Fares are collected at the front turnstile, so anyone boarding through the rear door rides for free. Checking it meant someone watching hours of video by hand, with no idea where to look.
Our Solution
A 120° overhead camera above the door and an on-board computer with an AI accelerator running a detection model trained to see people from above. A virtual line at the door counts entries only and filters double counts. Each entry creates an event with its clip, which travels to AWS (IoT Core, Lambda, DynamoDB and S3) and shows up in a dashboard where the operator watches the clips and marks each event as correct or not. Each device is configured remotely and the whole infrastructure is defined as code with Terraform.
Results Achieved
Piloting on a bus in service. In its first phase it ran 25 days straight without losing data, with video for 93% of events. In the current version, the device analyzes each frame in about 45 ms and a 15-second clip uploads in seconds once the bus reaches the depot. The operator goes from reviewing hours of footage to reviewing only the clip of each entry, and no alert is taken as fact until a person has watched the clip. We do not publish figures we have not measured.
Explore More Projects
View All ProjectsReady to Start?
Looking for the best technological solution?
We transform your ideas into innovative digital solutions that drive your business growth.
Contact us on WhatsApp!