• State TI Thinks Resolved
  • Locked Locked
  • Replies 3 replies
  • Answers 1 answer
  • Subscribers 100 subscribers
  • Views 464 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

TDA4VM: Sfm on fish-eye cameras & Valet parking application - PTK - Queries

John H. Gardner
Guru 53401 points
Part Number: TDA4VM


We received the following from one our customers.   Could you please clarify on below points for SfM on fish-eye camera application available in PTK demos, provided in the PSDKRA 6.2:

 

1.      Which camera model is used for fish-eye(FE) cameras in Sfm application, perception toolkit (PTK)

2.      Is it possible to disable INS/localization module and execution for a custom localization module

3.      What fps shall be achieved for 4 FE cameras, say, each image size is 1280x720 pixels.

 

Effectively, they would like to know how to configure valet parking demo application to execute using only vision sensor i.e., using 4 fish-eye cameras.

  • 0
    TI__Guru 53401 points

    Did hear back from the apps team on this.  Please see the following:

    John H. Gardner said:
    1.      Which camera model is used for fish-eye(FE) cameras in Sfm application, perception toolkit (PTK)

    The SFM camera App has been tested with just one camera with the camera should be facing the side and not front (Camera sensor is OV10635). The algorithm does not work well with the front facing cases.

     

    John H. Gardner said:
    2.      Is it possible to disable INS/localization module and execution for a custom localization module


    Our implementation needs INS for the location information {x,y,z,pitch,roll,yaw}. If this information is coming from somewhere else then it should be mapped to our data structures.


    John H. Gardner said:
    3.      What fps shall be achieved for 4 FE cameras, say, each image size is 1280x720 pixels.


    We have not tested the SFM/Valet Parking Apps on J7 EVM with live sensors, and only tested it on the TDA2x mounted on our car. Currently, we do not have plan in place to fix the J7 EVM in the car and test the SFM or Valet parking App with live sensors.

     

  • 0 in reply to John H. Gardner
    Prodigy 160 points

    Thanks for the response.

    Please let me know what is the fps achieved using OV10635 sensor? 

  • 0 in reply to Naveen KURUBA
    TI__Expert 3040 points

    Even though we have tested only 1 camera, I think it can process 2x720p@30 fps. (left and right cameras). For 4 cameras, we may need further optimization. Since the complexity mainly depends on the number of points, we could improve performances by smartly reducing the number of points to SfM.