Vehicle Occupancy Detection using Thermal Sensors

firefox

Objective

To develop a method for detecting the occupancy of vehicles (number of passengers) using a thermal camera.

Outcome

A complete system capable of high accuracy vehicle occupancy detection using thermal images.

Challenges

Run in real-time
Low space and computation requirements to be able be ported to a dedicated image processing chip
Perform reliably to be released as a consumer product

Core Model Architecture

Core model architecture highlighting the feature extraction. The last feature map is processed in two lanes. One for classification, and another performs deconvolution for blob detection.

The system captures raw thermal data from a thermal sensor, and evaluates captured thermal data to detect the number of passengers in the vehicle depending on the human body temperature.