How Perception Software Drives the Autonomous Vehicle Industry

  • July 30, 2024

What is Perception?

Perception is both a complex and natural process for humans. A necessary and fundamental part of how we interact and communicate with the world around us, perception helps us select, organize, and interpret information. Without this ability, engaging with the world would be an inefficient and complicated process.

While the ability to perceive our surroundings is nearly unnoticeable to us, research has shown that it takes about 80ms for the retina to see and register a new image under normal light conditions. A photographic camera operates similarly where light conditions affect the ability to register and successfully capture an image. If conditions are not optimal, the results can be less than ideal.

Perception in Autonomous Driving

As humans, our ability to classify objects and react to them quickly comes so naturally, it is often only apparent to us when we have trouble doing so. Automated and autonomous vehicles, much like humans, also have a lot to account for when perceiving their surroundings.

When encountering obstacles and other elements on the road, developing an “understanding” of size, shape, position, speed, and other factors are important in detecting, measuring, and mapping surroundings. LiDAR sensors achieve this understanding by bouncing light pulses off of surrounding objects and calculating the distance.

Beyond sensing technologies like LiDAR, autonomous vehicles require perception software that uses technical features superior to humans to achieve peak performance and safety capabilities. Pixel collision classification, for example, is important for autonomous and automated vehicles to accurately identify drivable areas and classify objects in 3D environments as collision-relevant or non-collision-relevant with pixel-level data accuracy. There are even two subclasses — “objects” (trucks, cars, bicycles, etc.) and “obstacles” (debris, tires, etc.) within the primary classification. It’s important for autonomous and automated vehicles to know the difference between a bicycle and a tire, for example, so they know if they can proceed safely.

Even a challenge like calibration that would not be an issue for humans is addressed by maintaining alignment between the LiDAR sensor’s coordinate system and the vehicle’s coordinate system. The performance of automated driving technologies cannot be impacted by vibrations typical of driving on rough terrain.

Innoviz and Autonomous Perception

Innoviz’s perception software leverages proprietary state-of-the-art AI algorithms and works in tandem with Innoviz LiDAR sensors, such as the InnovizOne, to provide leading detection, classification, tracking, and calibration capabilities.

Innoviz’s cutting-edge Perception Software, working in tandem with the InnovizTwo LiDAR sensor, delivers exceptional object detection and classification capabilities crucial for advanced autonomous driving systems. This sophisticated software processes the high-resolution 3D point cloud data from the InnovizTwo sensor, enabling it to detect and accurately identify a wide range of objects, including small and vulnerable road users such as cyclists, motorcyclists, and pedestrians.

The software’s advanced algorithms segment each frame in real-time, categorizing detected objects as either collision-relevant (obstacles or objects that may pose a risk) or not-collision-relevant (such as drivable ground surfaces or overhead structures). To ensure maximum accuracy, the Perception Software calibrates each individual frame, adjusting for environmental conditions and sensor positioning. With an impressive detection range of up to 300 meters, the system can identify and track an unlimited number of distinct objects per frame, providing a comprehensive and detailed understanding of the vehicle’s surroundings. This combination of long-range detection, high object capacity, and precise segmentation makes Innoviz’s Perception Software a powerful tool for enabling safe and reliable autonomous driving functionalities.

Perception Software on the Road

BMW has integrated Innoviz’s advanced Perception Software into its flagship i7 model, enabling Level 3 autonomous driving capabilities on German highways as of March 2024. This cutting-edge system utilizes Innoviz’s high-resolution LiDAR sensors and sophisticated AI algorithms to create a detailed 3D map of the vehicle’s surroundings in real-time. The Perception Software processes this data to accurately identify and track other vehicles, pedestrians, and potential obstacles, allowing the i7 to navigate complex traffic situations safely and efficiently. BMW’s implementation of this technology marks a significant step forward in the automotive industry, offering drivers the ability to take their hands off the wheel and eyes off the road in certain highway conditions, while the vehicle assumes full control of steering, acceleration, and braking.

The collaboration between BMW and Innoviz showcases the potential of LiDAR-based perception systems in enabling higher levels of autonomy in production vehicles. By leveraging Innoviz’s expertise in LiDAR technology and perception software, BMW has positioned itself at the forefront of autonomous driving innovation, setting a new standard for safety and convenience in the luxury automotive segment. As the rollout continues, this partnership is expected to pave the way for broader adoption of Level 3 autonomous features across BMW’s vehicle lineup, potentially revolutionizing the driving experience for consumers in Germany and beyond.

Learn More About Perception Software

Reach Out to Our Team Today

Innoviz team in the US Innoviz team in the US

Contact Us Now

(Pictured: the Innoviz team during their US roadshow showcasing the Perception Software in action)