• State TI Thinks Resolved
  • Locked Locked
  • Replies 8 replies
  • Answers 2 answers
  • Subscribers 100 subscribers
  • Views 2210 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.

TDA2EVM5777: object detection

swati N
Intellectual 875 points
Part Number: TDA2EVM5777


Hi ,

When I run the object detection usecase I could see "objtag" value as 465,458 like that ,as per classification user guide  the "objtag is unique ID associated with the object".

Kindly let me know where can I find these values and what basis these values are assigned.

Thanks,

Swati

  • 0
    TI__Mastermind 42515 points
    Hi Swati,

    in short "objtag" is unique ID which is assigned to a detected object and is used in case of object tracking.
    I have also pinged object detection expert to give additional details.

    Regards,
    Yordan
  • 0 in reply to Yordan Kamenov
    TI__Genius 10365 points

    Hi Swati,

    "objTag" is a 32bit field where bits 0-23 simply contains the track ID/track number and bits 24 to 31 contains the object type. The object type is an enum (TI_OD_ObjectType) in iobjdet.h header file in the inc folder.

    This field is better viewed in HEX format. 

  • 0 in reply to Shyam Jagannathan
    Intellectual 875 points

    Hi Shyam,

    Thanks for the reply.

    We are trying to use the existing object detection usecase to detect object like barricades, but  the object detection algorithm

    detects pedestrian ,vehicle and traffic signal recognition only.

    In the code I could see the "TI_ OD_ObjectType" has only pedestrian, vehicle and TSR.

    Please let me know how can I use existing algorithm for detecting other objects.

    Under  "TI_ OD_ObjectSubType" structure their is a member "Ti_CLASS_OTHER" can you please tell me why this member is used for??

    Thanks,

    Swati

  • 0 in reply to swati N
    Intellectual 875 points
    Hi,

    Can you please reply for my above query ,I am trying to identify objects other than vehicle ,pedestrian , and traffic signal recognition .

    Thanks,
    Swati
  • 0 in reply to swati N
    TI__Expert 4885 points
    Hi, ,

    The expert is notified about your new question.

    Regards,
    Mariya
  • 0 in reply to swati N
    TI__Genius 10365 points

    Hi Swati,

    Firstly apologize for a late reply, for some reason I didn't receive an e-mail alert on this thread. 

    The current version of Object Detect (v00.05) can only detect Pedestrians, Vehicles, Traffic Signs. The algorithm uses HOG and a 2-level symmetric AdaBoost Classifier assuming certain parameters. For detecting new class of objects such as barricades etc, you will have to freshly train by keeping similar HOG and Adaboost parameters and structures.

    The TI_OD_ObjectSubType is used for identifying the "type" of detected Traffic Sign. This field is populated by Object Classification module which runs a CNN classifier. The CNN Classifier is trained to identify 43 different types of German Traffic Signs although there are more in the real world. Any traffic sign other than these 43 will be classified as TI_CLASS_OTHER. 

    Regards,
    Shyam

  • 0 in reply to Shyam Jagannathan
    Intellectual 875 points
    Hi Shyam,

    Thanks for the reply.
    can you please tell me how training of objects are done in TI ,suppose if I want to detect a pedestrian with no head how to train this image .
    please let me know what tool I need to use in TI to train the objects.

    Thanks,
    Swati
  • 0 in reply to swati N
    TI__Genius 14090 points
    Hi Swati,

    You can refer online references about training a classifier. We will be very happy to help you in device specific questions and questions about our sofftware. But would expect our partenrs to do value add in this space (liek training, functional defination of an algorithm) which is completly outside the scope of our device.

    I will check with some of my collegeues to see if they can give you some pointers. But as I said you should be able to find plenty of references online and in academia about this

    Pramod