SPRACX2 April 2021 TDA4VM , TDA4VM-Q1
This application runs on saved images and INS data, and therefore skips the camera sensor to VPAC leg of the diagram above. Documentation on how to run the application as well as all the necessary libraries can be found in the SDK, and are listed in Section 9. Application specific documentation can be found here .
A sample output of this application is shown in Figure 7-1. In the top of diagram, we show the output of Dense Optical Flow and in the bottom the output of Sparse Optical Flow (SOF). From this diagram, it is clear why DOF is preferred over SOF for applications such as OG Mapping, as oncoming cars cannot be detected with a sparse point cloud. The experiment runs on simulated data generated using the Carla Simulator 0.9.9.4 [1], with a virtual camera mounted on the front of the vehicle. Here, free cells are marked green, occupied red and unknown yellow. This example runs at 30 fps on 0.5 MP camera data with less than 60% C7x DSP utilization to generate 20,000 3D key points. Although the ISP portion is skipped in this application, the ISP has the ability to process data at 600 MP/s and, therefore, can accommodate a 0.5 MP monocular camera at 30 fps. The user can request a Data Sheet with precise information about different core utilization here.
Examples in the SDK, including the SFM OG mapping example, are made available to customers to shorten their development time. If customers prefer their own pipeline or a modified pipeline, these examples can be used as reference or as a starting point.