Research Internship: Lightweight Metric 3D Head Detection in Dense Scenes
Purpose
The internship aims to advance the "Freeflow" technology, focusing on the detection and tracking part. Freeflow moves beyond traditional biometric gates—which require passengers to stop for facial, fingerprint, or iris scanning—by leveraging advanced video analytics to identify individuals seamlessly as they move. This creates a frictionless user experience by eliminating physical barriers.
We want to design a lightweight neural network that, from a single calibrated camera, detects (i.e., predicts its 2D image position) every pedestrian head in a possibly dense scene and predicts for each head its metric depth. Each detection is then lifted to a 3D point. The 3D points coming from several synchronized cameras are associated in 3D and fed to a 3D tracker. The final goal is to achieve more robust, 3D-aware tracking.
Key Missions
-
Analyzing and understanding the current crowd localization model architecture and performance.
-
Conducting experiments to evaluate tracking precision.
-
Collaborating with a team of researchers to contribute to the overall project goals.
-
Documenting findings and presenting results.
Profile & Other Information
Education
Final-year engineering student at a leading engineering school, or a second-year Master's student (M2) specializing in computer vision, image processing, or deep learning.
Technical Skills
-
Strong knowledge of computer vision and deep learning.
-
Proficiency in Python, OpenCV, and PyTorch.
-
Solid background in data analysis and software development.
Interpersonal Skills and Languages
-
Proficiency in written and spoken English, including the ability to read and understand scientific papers.
-
Curious, proactive, and self-motivated, with the ability to work independently.
-
Results-driven, with a strong problem-solving mindset.
-
Excellent communication skills, with the ability to present ideas clearly and persuasively.
-
Strong interest in contributing to the full lifecycle of an algorithm, from design and implementation to integration, testing, and optimization.
Location: La Défense, Courbevoie, France
Duration: 6 months
Expected Start Date: Approximately March 2027