• State
  • Locked Locked
  • Replies 4 replies
  • Subscribers 102 subscribers
  • Views 121 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

PROCESSOR-SDK-AM62A: Does it support traffic signal and light signal judgment?

Daniel Cheng1
Genius 5965 points
Part Number: PROCESSOR-SDK-AM62A
Other Parts Discussed in Thread: MIDAS

Tool/software:

Hi Expert,

Does SDK support traffic signal and light signal judgment?

Thanks

Daniel

  • 0
    TI__Mastermind 35140 points

    Hi Daniel, are you referring to have an OOB demo example for Traffic signal? If so, I am not aware of any, could you please clarify your use case?

    thank you,

    Paula

  • 0
    TI__Genius 10456 points

    Hi Daniel,

    Echoing what Paula said, we do not have this type of example application built into the SDK. It could be implemented (similar to a lane detection system) using the C7xMMA AI accelerator to run a neural network that is trained to recognize traffic signals/lights.

    BR,
    Reese

  • 0 in reply to Reese Grimsley
    Genius 5965 points

    Hi Reese

    AM62A is 2 TOPS vision SoC,

    2 TOPS AI have enough computing power to meet the following needs at the same time?

    If not, do you have any SOC suggestions?

    1.traffic signal and light signal judgment

    2.Lane Departure Warning System，LDWS

    3.Forward Collision Warning System (FCW)

    4.Monocular Depth Estimation

    Can distance judgment be achieved using the Raspberry Pi Camera connected to the EVB MIPI CSI default?

    dev.ti.com/.../node

    Can a single Raspberry Pi Camera meet the above situational requirements?

    Thanks

    Daniel

  • 0 in reply to Daniel Cheng1
    TI__Genius 10456 points

    Hi Daniel,

    This looks/feels like a front-camera use case. AM62A is capable of the types of models needed here, and can run multiple models that are simultaneously loaded.

    Running this many models concurrently at an acceptable framerate would need further investigation. It will depend strongly on the complexity of the chosen models and the resolutions they operate at. 

    Daniel Cheng1 said:

    Can distance judgment be achieved using the Raspberry Pi Camera connected to the EVB MIPI CSI default?

     i assume you found my depth estimation demo :) . This was using the MIDAS model, and can estimate distance from a single camera at around 60 FPS. Enabling the camera requires a device tree overlay (DTBO) be added and the model downloaded (depending on the SDK version, this may require recompilation). The demos has a subpage for this application that points to a Github repo with additional information. 

    Daniel Cheng1 said:

    Can distance judgment be achieved using the Raspberry Pi Camera connected to the EVB MIPI CSI default?

    Do you mean the IMX219 camera? It would probably not be the right choice for this type of application, but it would be possible. Most front-camera applications would prefer to use a device with HDR/WDR to better handle the variety of lighting / shadows that a FC application needs to work in.

    BR,
    Reese