Freespace Detection


Further Key Components

Autonomous Driving

We distinguish between short and long term goals in our development process. On the short run we want to enable new ADAS technologies and take existing ADAS to the next level with components like our Freespace Detection. By focusing on camera-based technologies we furthermore ensure to be able to provide low cost, mass-market solutions.
On the long run we want to use the components that proved to be reliable in the ADAS implementations and use these for Autonomous Driving. A key component like our Freespace Detection can thereby improve or replace expensive technologies like LiDAR that are currently mostly the choice for developing fully autonomous vehicles.
As a first glimpse into the ADAS technologies that could be enabled with our Freespace Detection already today we show a few use cases below.
Advanced Collision Avoidance
Current Path
Alternative Path
Current Collision Avoidance Systems are mainly based on Lane Detection and a separate Object Detector for predefined object classes. These systems lack the crucial ability to "see" objects of arbitrary classes that were not defined in the detector before. This can result in the Collision Avoidance System totally overseeing some obstacles. By using bounding box detectors they also lack to find the actual shape of the obstacle. In addition to this the Lane Detection approach struggles in scenarios where clear lane markings or clear lane borders are missing and thereby sometimes overseeing large areas of drivable space.
The Freespace Detection based Collision Avoidance approach overcomes these drawbacks. It recognizes arbitrary obstacles and their actual shape by looking for the free, drivable space. Therefore it is not limited to specific object classes and its bounding boxes. Moreover, because it does not rely on special lane boundaries to find free areas it can also find "escape zones" on any kind of street.
Parking Spot Detection
Parking Spot

The free street layout that is given by the Freespace Detection can be used to find possible parking spots in the near environment. Given the size and shape of the ego-vehicle a successive algorithm can check the freespace for suitable parking areas in the near environment. On top of that it can measure the size and shape of a potential parking spot and eventually plan a path for parking.

Autonomous Parking (Summon Function)

Already the exclusive use of Freespace Detection can enable autonomous driving abilities for a car to some extent. In areas like car parks where additional environmental informations are not relevant for save driving a car equipped with a Freespace Detection would be able to drive around that space by itself.

Nowadays current systems that enable such a feature mainly use ultrasonic parking sensors. These sensors are only able to give a shallow representation of the drivable space in the very near environment. This results in the car not being able to properly plan its path over a longer distance which furthermore results in very slow driving and bad path planning decisions.

Our camera based Freespace Detection on the other hand gives a very detailed representation of the drivable space even in very far areas. This enables smart path planning decisions and also a higher driving speed for e.g. a Summon Function.

Free 360° Surround View

Modern cars use multiple cameras around the car to enable a feature called Surround- or Top-View. With our Freespace Detection not requiering stereo camera images the detection can be applied to each of these monocular cameras which then enables a 360° Freespace Detection.