Bosch Designs and Implements a Lidar Point Cloud Classifier with MATLAB and Deep Learning Toolbox

Support from MathWorks engineers helped to accelerate the entire project, from faster realization of deep learning models to the collaborative realization of a multitarget workflow.

Key Outcomes

  • Established framework-independent deep learning platform
  • Accelerated development of accurate point cloud classifier
  • Generated and deployed embedded code

At Bosch Global Software Technologies (BGSW), engineering teams are developing advanced driver assistance systems (ADAS). These systems often require processing large volumes of real-time data from multiple sensors using on-vehicle computers. One BGSW team has developed a Lidar Development Kit (LDK) to accelerate the development and deployment of ADAS applications that employ deep learning networks to process lidar data.

The team chose MATLAB® as the basis for LDK to meet two key requirements: framework independence and deployment interoperability. Rather than limiting the LDK to one specific deep learning framework—such as TensorFlow™, PyTorch®, Keras, and ONNX—the team wanted a platform and a workflow that supported all of them. With MATLAB and Deep Learning Toolbox™, the team can import a model—one that that classifies lidar point cloud data, for example—from any of these frameworks, customize it as needed, and then retrain it. Once the model is validated and finalized in MATLAB, the team can then use MATLAB Coder™ or Embedded Coder® to generate C/C++ code for deployment to a variety of hardware platforms and target processors.