How to take photo (How to save frame image)

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU) Jetson Nano
• DeepStream Version 5.1
• JetPack Version (valid for Jetson only) 4.5.1
• TensorRT Version 7.1.3+cuda10.2
• Issue Type( questions, new requirements, bugs) Question

Hi,
I modified deepstream_test1_usb.py little. I wonder how can I take pictures with detection.

Thanks

deepstream_test_1_usb_2.py (17.0 KB)

3

Can you elaborate your requirement? You want to save the frame image which some objects are detected?

4

I want to save frame image when person and helmet count not equal.

5

There is image save sample deepstream-imagedata-multistream.

6

Thank you. I will check.

7

Hi,

I edit my code like below.

            if obj_counter[PGIE_CLASS_ID_PERSON] != obj_counter[SGIE_CLASS_ID_HELMET]:
                track_id = obj_meta.object_id
                now = time.time()
                if flag==0 and track_id != last_track_id:
                    #print(str(track_id) + ' -- ' + str(last_track_id))
                    n_frame = pyds.get_nvds_buf_surface(hash(gst_buffer), frame_meta.batch_id)
                    frame_copy = np.array(n_frame, copy=True, order='C')
                    frame_copy = cv2.cvtColor(frame_copy, cv2.COLOR_RGBA2BGRA)
                    last_track_id = track_id
                    save_image = True
                    print('Photo')
                    takenow = time.time()
                    flag = 1
                    TAKE_PHOTO = True
                if now>takenow+5:
                    print('Reset')
                    flag = 0
            else:
                TAKE_PHOTO = False
            try: 
                l_obj=l_obj.next
            except StopIteration:
                break
        if save_image:
            img_path = "{}/stream_{}_frame_{}.jpg".format(folder_name, frame_meta.pad_index, frame_number)
            print(img_path)
            cv2.imwrite(img_path, frame_copy)

Its works like I wanted.

But I had problem with orginal img_path provided in example below. frame_{} folder wasnt created. I don’t know if I missed the part for that folder creation but I changed it to stream_{}_frame{}

            img_path = "{}/stream_{}/frame_{}.jpg".format(folder_name, frame_meta.pad_index, frame_number)

Thank you

8

hello I have tried using deepstream-imagedata-multistream but for me it shows bus error , segmentation fault after saving 1-2 frames . If I comment n_frame = pyds.get_nvds_buf_surface(hash(gst_buffer), rame_meta.batch_id)
Then there is no probelm arising, could you help me to figure out this problem. I have been stuck with it for long . I have tesla k40 gpu, and all other applications are working fine except when I try to extract frame meta…

9

Hey @bhatiyaarpit95 . I am sure if you open new topic you can find faster solution.

But you missed ‘f’ in rame_meta below.

n_frame = pyds.get_nvds_buf_surface(hash(gst_buffer), rame_meta.batch_id)

I am not sure if this is the cause of your problem but you can try.

Edit: I saw yor py file here. It seems ok to me.

Thanks

10

Thanks for your time, yes that ‘f’ missing is not problem btw… Did it worked fine in your system.

11

It works fine but I didn’t use source _bin and draw_bounding_boxes. Problem might be one of them