Sigtry

Why do we make Fourier transform to signals? The first thing that came to my mind for this question is that the Fourier transform has many nice features, it can transfer complex convolution in the time domain into a simple product in the frequency domain, it can simplify the filter design in frequency domain, its linearity, symmetry property etc. But those nice features can’t explain well why we need it but not other orthogonal expansions. While I’m looking for the answer to this question, my understanding of Fourier expansion gets refreshed and also extended to some applications of signal expansion in other fields.

In last article, the general data processing chain inside a Cyber Component is introduced in detail. We know that the customized iniitalization and data processing of a perception component (for example the Lidar detection component) will be implemented in its Init() and Proc() member functions. In this article I would like to make the perception process clear, but won't illustrate the algorithm details.

People will want to see where are the deep learning models being invoked in perception at first, actually there are several levels till it appears. Like the last article, I will begin from a dag file where to launch the perception. Under /modules/perception/production/dag path there are several dag files which will combine different perception components into a module. Each perception component class is a child class of Cyber::Component. All perception components are defined under /modules/perception/onboard path. Here will take dag_streaming_perception_lgsvl.dag as example, which includes the following perception components:

• LidarDetectionComponent
• LidarTrackingComponent
• RadarDetectionComponent
• MultiSensorFusionComponent

And at the time of writing, the code structure of the perception process for different sensors has not been unified yet, especially radar perception part. I will address them separately.

Recently, I'm learning Apollo project which is an open source project for autonoumous driving. At beginning I hoped I can focus on only the perception module, but soon I realized it's not possible to go without touching the bottom layer -- Cyber RT framework. At the time when I'm learning now, there was not much info about the code strucuture and process chain can be found from the official documents in the project. This blog is my summary on this(version 8.0). I will put it to the following parts: - Important concept items in Cyber component - Initialization of a Cyber component and data receiving/processing inside - Data transimit in Cyber framework - Perception module break down