In this presentation, we show an approach for 3D object tracking using front cameras. This involves object classification, localization, velocity estimation and trajectory estimation. 3D inference using a single camera is an ill-posed problem due to the absence of depth information. Therefore, a stereo camera setup is used, along with geometric and probabilistic algorithms to estimate the depth and velocity. The system is divided into two layers. The first extracts the 2D features of the observed objects, such as their 2D bounding boxes on the images. The second uses those 2D features to extrapolate the 3D properties of the objects.

• Reliable 3D object tracking is crucial for ADAS systems
• Low-cost camera-based geometric methods can be used for perception
• Camera-based object tracking is still a topic of research