SwissMoves – WIM

SwissMoves – WIM Study about truck traffic on swiss highways

  • HES-SO Fribourg
  • Prof. Marc-Antoine Fénart
  • Prof. Beat Wolf
  • Dr. Oussama Zayene
  • Mr. Maël Vial
  • Image analysis
  • Object detection
  • Data analysis
External mandate
01.01.2022 – 01.06.2023

In this study we analyse video data from Swiss highways with custom object detection models to identify the types of trucks on a certain highway strip. This study is a part of a running project in partnership with the OFROU, and represents a joint collaboration between iCoSys and iTEC institutes. The following figure provides more details about the context and purpose of the study.

The study involves several steps ranging from material installation and data storage to smart data analysis using the state-of-the-art real-time YOLO model. The general pipeline is depicted in the figure below.


A custom dataset has been created to train and evaluate machine learning models. The dataset includes 11 different classes of heavy vehicles. A total of 3680 video frames representing these classes, have been extracted from 70 hours of raw video, and then annotated in a semi-automatic way. We used the dataset to realise more than 50 training and inference experiments. The figure below shows some results of truck classification.

YOLO performs well and gives very interesting classification results despite the presence of several challenges like motion, shadow and inter-class similarity.