Robust environment perception is a prerequisite for automated driving. It allows to perceive and safely interact with other traffic participants such as cars and pedestrians. Also, traffic signs and traffic lights have to be detected to infer traffic rules.
Object detection and tracking
We employ state-of-the-art neural networks for object detection. We not only provide robust bounding boxes with semantic class information but we also track them temporally which allows to predict object behavior.
Pixel-wise semantic labels allow us to understand the scene around us. Which areas are we allowed to drive in and where will pedestrians most likely walk? All these questions can be answered if we know our surrounding.
Monocular depth estimation
As an alternative to stereoscopic vision and LiDAR, learned monocular depth can be used to estimate the 3D location of objects.
There are many free and open source solutions on the internet. However, they will most likely not work well with your sensor setup and in the environment you intend to use them. Also, they either run too slow on your system or they don’t make use of its full potential. Therefore, we offer our customers neural networks that are custom tailored for their needs and specifically adapted to their hardware. If you wish to integrate more classes or training examples later on we will take care of the whole process from data labeling to retraining.