Efficient synchronization and processing: Application of ADTF streaming service in autonomous driving data acquisition

With the development of autonomous driving technology, the intelligence of vehicles continues to improve, which reflects the vehicle's perception, decision-making and execution capabilities. In the process of algorithm development and iteration, the key to improving testing and development efficiency lies in the high-quality collection of multi-sensor data to ensure the synchronization, integrity and consistency of data.

To address this challenge, an effective data integration strategy must be adopted. This includes developing components that handle different data rates and formats, and designing synchronization mechanisms that precisely align data in time. This creates a unified data stream format that detects sensor observations in real time and stores them on disk.

To solve the above problems, ADTF provides a powerful Streaming Service, which is specially designed for the needs of multi-sensor data collection for autonomous driving. With its efficient data processing capabilities, flexible architecture design and powerful synchronization mechanism, ADTF Streaming Service provides a high-quality solution for autonomous driving data acquisition systems.

1. ADTF Streaming Service

In ADTF, streaming services play a vital role, defining the entry point (Streaming Source) and exit point (Streaming Sink) of the system. Specifically, streaming services can be both the starting point and the end point of a data pipeline, and they are usually used to process sample data and data triggers from hardware.

1. Streaming Source

The stream service source is the entry point for samples and data triggers into the system. Typically, any device link is implemented as a stream service source, and the components are shown in Figure 1 below:

Figure 1: Streaming Source components

Streaming Source supports the following application scenarios:

(1) Read video streams from cameras
(2) Read CAN messages from CAN bus devices
(3) Act as a hard disk reader to provide file-based simulation data
(4) Receive samples from distributed systems such as ROS (Robot Operating System) or FEP (Function Engineering Platform) via a network or inter-process connection

2. Streaming Sink

The stream service sink is the exit point for samples and triggers to leave the system. Typically, any device link will be implemented as a stream service sink. The components are shown in Figure 2 below:

Figure 2: Streaming Sink Component

Streaming Sink supports the following application scenarios:

(1) Write raw CAN messages to CAN bus devices
(2) Write FlexRay cycles or Automotive Ethernet PDUs to devices
(3) Create hard disk access for file-based data logging and high-performance recording

2. Data Link

Streaming service is the key to achieving efficient data processing and transmission. Through Streaming Source and Streaming Sink, corresponding data links can be built according to different actual application needs, including data pipes, substreams, and trigger pipes.

1. Data Pipe

The data pipeline is a bridge connecting the sample writer (Streaming Sink) and the sample reader (Streaming Source). In ADTF, as shown in Figure 3, a common data pipeline starts from the output pin (Out Pin) of the sample writer of the send filter, through a sample stream (Sample Stream), to one or more input pins (In Pins) and their corresponding sample readers.

Figure 3: Data Pipe

2. Substreams

Subflow is a way to reduce the complexity of pins and connections in the filter graph, as shown in Figure 4. By using subflow, the data flow path can be simplified, improving the readability and maintainability of the system.

Figure 4: Subflow

3. Trigger Pipe

The trigger pipeline is a connection based on the trigger path in ADTF, usually starting from the Active Timer Runner, as shown in Figure 5, which triggers the connected components, similar to the runner of the filter.

Figure 5: Trigger pipeline

For example, in autonomous driving data collection, by designing a flexible trigger mechanism, the data pipeline can synchronize the data streams from radar, camera, and lidar to ensure their temporal alignment.

Conclusion

The key to the application of ADTF Streaming Service in the field of autonomous driving data collection is its ability to achieve efficient synchronization and processing of multi-sensor data. The following are several key application points:

Multi-sensor data fusion: Synchronize data from different sensors through data pipelines to achieve high-precision environmental perception.

Real-time decision support: Use trigger pipelines to quickly respond to events triggered by sensors and provide real-time data to decision algorithms.

Data storage and playback: Efficient data storage is achieved through Streaming Sink, and data playback and analysis are performed through sub-streams.

ADTF Streaming Service provides a powerful tool for autonomous driving data acquisition systems that can not only process and synchronize multi-sensor data, but also adapt to changing development needs.