AM68A: How to trigger / print message when detect person

Hi,

Yes, this should be possible to do on the EVM.

Thank you,

Fabiana

HOW IS POSSIBLE

please explain the steps 

Hi Mohammed Niyas,

We do not have any tutorial for this, but you can take a look at how the example edgeai-gst-apps code to see how this can be implemented in a similar application. The python-based application is located in the /opt/edgeai-gst-apps/apps_python directory.

Thank you,

Fabiana

HI fabiana,
I have completed the following steps using C code:

1. Captures live video from a camera.
2. Processes the video through a GStreamer pipeline that includes TI’s inference elements for object detection.
3. Records the stream into 10-second MKV files.
4. Manages recorded files by retaining only the last three segments, deleting older ones.
5. Listens for user input, so when the user presses 'y', it copies the current, previous, and next segments.

Now, I need to trigger an action when a person is detected. Is it possible to implement this trigger using C code, or alternatively, how can I extract metadata related to object detection? 

gchar *pipeline_str = g_strdup_printf(
"v4l2src device=/dev/video-usb-cam0 io-mode=2 ! image/jpeg, width=1280, height=720 ! jpegdec ! tiovxdlcolorconvert ! video/x-raw, format=NV12 ! "
"tiovxmultiscaler name=split src_0::roi-startx=0 src_0::roi-starty=0 src_0::roi-width=1280 src_0::roi-height=720 target=0 ! "
"queue ! video/x-raw,width=320,height=320 ! "
"tiovxdlpreproc model=/opt/model_zoo/TFL-OD-2020-ssdLite-mobDet-DSP-coco-320x320 out-pool-size=4 ! application/x-tensor-tiovx ! "
"tidlinferer target=1 model=/opt/model_zoo/TFL-OD-2020-ssdLite-mobDet-DSP-coco-320x320 ! post.tensor "
"split. ! queue ! video/x-raw,width=480,height=480 ! post.sink "
"tidlpostproc name=post model=/opt/model_zoo/TFL-OD-2020-ssdLite-mobDet-DSP-coco-320x320 alpha=0.4 viz-threshold=0.6 top-N=5 display-model=true ! "
"tee name=video_tee ! queue ! v4l2h264enc ! h264parse ! queue ! mux. "
"alsasrc device=hw:0,0 ! audio/x-raw,format=S16LE,rate=16000,channels=1 ! audioconvert ! audioresample ! "
"avenc_aac bitrate=128 ! aacparse ! queue ! mux. "
"matroskamux name=mux ! filesink location=%s sync=true",
output_file
);

How extract the class ID and name from the tidlinferer element

Hi Mohammed Niyas,

This is not a feature in our sample applications that can be enabled, so you will have to implement this yourself. If you take a look at either the c++ or python GStreamer application, you can see where this data lies and make the changes required to achieve this. Please let me know if you have specific questions about our code.

Thank you,

Fabiana

What about OptiFlow? From what I observed, the apps_cpp and apps_python examples use OpenCV for post-processing instead of TIDL elements. However, our requirement is to use GStreamer, where tidlpostproc is utilized for post-processing.

I also noticed that OptiFlow runs an end-to-end GStreamer pipeline when executing the demo. Our main goal is to achieve the same using GStreamer while extracting either the class name or at least the class ID, which we can then map to the COCO dataset. tidlpostprocess will have class names any way to access this.

When I tried running the C++ application (apps_cpp), I encountered a graph-related error. I previously faced this issue but couldn't resolve it at the time.

root@am68a-sk:/opt/edgeai-gst-apps/apps_cpp# ./bin/Release/app_edgeai ../configs/image_classification.yaml

Number of subgraphs:1 , 34 nodes delegated out of 34 nodes

APP: Init ... !!!
12257.155477 s: MEM: Init ... !!!
12257.155556 s: MEM: Initialized DMA HEAP (fd=6) !!!
12257.155752 s: MEM: Init ... Done !!!
12257.155779 s: IPC: Init ... !!!
12257.197586 s: IPC: Init ... Done !!!
REMOTE_SERVICE: Init ... !!!
REMOTE_SERVICE: Init ... Done !!!
12257.206213 s: GTC Frequency = 200 MHz
APP: Init ... Done !!!
12257.206971 s: VX_ZONE_INFO: Globally Enabled VX_ZONE_ERROR
12257.207203 s: VX_ZONE_INFO: Globally Enabled VX_ZONE_WARNING
12257.207236 s: VX_ZONE_INFO: Globally Enabled VX_ZONE_INFO
12257.210168 s: VX_ZONE_INFO: [tivxPlatformCreateTargetId:134] Added target MPU-0
12257.210712 s: VX_ZONE_INFO: [tivxPlatformCreateTargetId:134] Added target MPU-1
12257.210892 s: VX_ZONE_INFO: [tivxPlatformCreateTargetId:134] Added target MPU-2
12257.212165 s: VX_ZONE_INFO: [tivxPlatformCreateTargetId:134] Added target MPU-3
12257.212225 s: VX_ZONE_INFO: [tivxInitLocal:126] Initialization Done !!!
12257.212260 s: VX_ZONE_INFO: Globally Disabled VX_ZONE_INFO
12257.229918 s: VX_ZONE_ERROR: [ownContextSendCmd:912] Command ack message returned failure cmd_status: -1
12257.230313 s: VX_ZONE_ERROR: [ownNodeKernelInit:604] Target kernel, TIVX_CMD_NODE_CREATE failed for node TIDLNode
12257.230442 s: VX_ZONE_ERROR: [ownNodeKernelInit:605] Please be sure the target callbacks have been registered for this core
12257.230531 s: VX_ZONE_ERROR: [ownNodeKernelInit:606] If the target callbacks have been registered, please ensure no errors are occurring within the create callback of this kernel
12257.230638 s: VX_ZONE_ERROR: [ownGraphNodeKernelInit:690] kernel init for node 0, kernel com.ti.tidl:1:1 ... failed !!!
12257.230696 s: VX_ZONE_ERROR: [ TIDL subgraph MobilenetV1/Predictions/Reshape_1 ] Node kernel init failed
12257.230794 s: VX_ZONE_ERROR: [ TIDL subgraph MobilenetV1/Predictions/Reshape_1 ] Graph verify failed
TIDL_RT_OVX: ERROR: Verifying TIDL graph ... Failed !!!
TIDL_RT_OVX: ERROR: Verify OpenVX graph failed
graph
[07:07:56.000.000000]:ERROR:[startPipeline:0141] gst_element_set_state() failed.
[07:07:56.000.000085]:ERROR:[setupFlows:0250] Failed to start GST pipelines.
terminate called after throwing an instance of 'std::runtime_error'
what(): EdgeAIDemoImpl object creation failed.
Aborted (core dumped)

Hello,

What SDK version are you using? Have you made any changes to the configuration file or anything else? Do you have a display connected to the board?

Thank you,

Fabiana

SDK version 10.0.1, no i have not made any changes to the configuration file.
Thanks connecting monitor has worked the issue is no more.

if it is possible can you please answer this.

Thank you very much for your reply. I ran the C++ application and modified the object_detection.yaml file to use a USB camera for detection.

I attempted to add a print statement in /common/src/post_process_image_object_detect.cpp to display the detected object's name in the terminal. However, it seems the print statement is not working. This might be due to the existing logs generated during the execution of the object detection application, which could be overwhelming the terminal output and preventing our custom print statement from appearing.

Additionally, I tried to inspect the script being executed, located in bin/release/app_edgeai, but it is not in a readable format, making it difficult to debug further.

Could you please help me understand why the print statement isn't appearing in the terminal and suggest a way to resolve this issue?

Hi Nihal,

Did you rebuild after making the change to the C++ application? C++ apps can be modified and built on the target using below steps.

/opt/edgeai-gst-apps/apps_cpp# rm -rf build bin lib
/opt/edgeai-gst-apps/apps_cpp# mkdir build
/opt/edgeai-gst-apps/apps_cpp# cd build
/opt/edgeai-gst-apps/apps_cpp/build# cmake ..
/opt/edgeai-gst-apps/apps_cpp/build# make -j2

Thank you,

Fabiana

I appreciate your help! I was able to print the object name successfully. I also implemented logic to assign a variable and print only when a person is detected exclusively.

Now, I want to record a 10-second video three times in a FIFO manner—meaning when a new recording starts, the oldest file should be deleted and replaced with the latest one. I’ve done this before using splitmuxsink, but I’m unsure how to implement it in this case.

Currently, I can save and stream video simultaneously, but I need help achieving continuous recording while maintaining only the last three files.

Below is the relevant section from edgeai_demo_config.cpp where the video is saved:

else if (sinkType == "video")
{
    string h264enc = gstElementMap["h264enc"]["element"].as<string>();
    string encoder_extra_ctrl = "";

    if (h264enc == "v4l2h264enc")
    {
        encoder_extra_ctrl = "controls"
                             ",frame_level_rate_control_enable=1"
                             ",video_bitrate=" + to_string(m_bitrate) +
                             ",video_gop_size=" + to_string(m_gopSize);

        m_gstElementProperty = {{"extra-controls", encoder_extra_ctrl.c_str()}};
    }

    makeElement(m_dispElements, h264enc.c_str(), m_gstElementProperty, NULL);

    // Add H.264 Parser
    makeElement(m_dispElements, "h264parse", m_gstElementProperty, NULL);

    // Add MKV Multiplexer
    makeElement(m_dispElements, "matroskamux", m_gstElementProperty, NULL);

    // Configure filesink to store in output.mkv
    m_gstElementProperty = {{"location", "output.mkv"},
                            {"name", name.c_str()}};

    makeElement(m_dispElements, "filesink", m_gstElementProperty, NULL);
}

Additionally, in my object_detection.yaml, I’ve created two output flows:

outputs:
    output0:
        sink: kmssink
        width: 1280
        height: 720
        overlay-perf-type: graph
    output1:
        sink: /opt/edgeai-test-data/output/output_video0.mkv
        width: 1280
        height: 720
    output2:
        sink: /opt/edgeai-test-data/output/output_image_%04d.jpg
        width: 1920
        height: 1080
    output3:
        sink: remote
        width: 1920
        height: 1080
        port: 8081
        host: 127.0.0.1
        encoding: jpeg
        overlay-perf-type: graph

flows:
    flow0: [input1, model2, output0]
    flow1: [input1, model2, output1]

Here, input1 is a USB webcam video, and model2 is YOLOX.

Could you help me implement continuous recording in 10-second intervals while ensuring only the last three video files are retained, following a FIFO approach?

Ok Thanks