ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question

Real time Image topic subscription

asked 2018-05-16 12:17:28 -0600

erivera1802 gravatar image

Hi everyone, im trying to work on an object detector in ROS, using the deep learning modules from OpenCV. My problem is related to the velocity of the node. The camera node publish at 20 hz, and when I do not activate the deep learning part, my node publish equally fast. When I activate the deep learning functions, the rate drops to 15 hz, so it shouldnt be a great problem.

However, when I look at the results (Im working over a Rosbag), what I see, is that the Image that is published and edited by my node, comes like 4 seconds later than the original image. What I mean I want to do, is that the node, no matter if it has to skip frames, uses the actual frame he becomes from the publishing topic, the result of it would be a not so smooth sequence of images, but it would be showing the present time. What I suspect its happening now, its that the node is just picking the next image that comes from the camera topic, like it is being saved on a buffer.

This is my code:

import rospy
from std_msgs.msg import Header
from dynamic_reconfigure.server import Server
import cv2 as cv
import numpy as np
from cv_bridge import CvBridge,CvBridgeError
from tensorflow_ros_tool.cfg import ObjectConfig
from tensorflow_ros_tool.interface.ObjectInterface import ObjectInterface
class Object:
def callback_subscriber(self, msg):
    Called when a new message arrives
    # do your stuff here

    cv_image = self.bridge.imgmsg_to_cv2(msg, desired_encoding="passthrough")
    now = rospy.get_rostime()

    self.cvNet.setInput(cv.dnn.blobFromImage(cv.resize(cv_image, (300, 300)), 0.007843,(300, 300), 127.5))

    cvOut = self.cvNet.forward()
    CLASSES = ["background", "aeroplane", "bicycle", "bird", "boat",
               "bottle", "bus", "car", "cat", "chair", "cow", "diningtable",
               "dog", "horse", "motorbike", "person", "pottedplant", "sheep",
               "sofa", "train", "tvmonitor"]
    COLORS = np.random.uniform(0, 255, size=(len(CLASSES), 3))

    for i in np.arange(0, cvOut.shape[2]):
        confidence = cvOut[0, 0, i, 2]
        if confidence > 0.3:
            idx = int(cvOut[0, 0, i, 1])
            box = cvOut[0, 0, i, 3:7] * np.array([960, 540, 960, 540])
            (startX, startY, endX, endY) = box.astype("int")
            label = "{}: {:.2f}%".format(CLASSES[idx], confidence * 100)
            #print("[INFO] {}".format(label))
            cv.rectangle(cv_image, (startX, startY), (endX, endY),COLORS[idx], 2)
            y = startY - 15 if startY - 15 > 15 else startY + 15
            cv.putText(cv_image, label, (startX, y),
            cv.FONT_HERSHEY_SIMPLEX, 0.5, COLORS[idx], 2)

    self.interface.dummy_publisher.publish(self.bridge.cv2_to_imgmsg(cv_image, "bgr8"))
    now = rospy.get_rostime()-now
def reconfigure_callback(self, config, level):
    Called when someone reconfigures this node (or at startup)
    return config

def __init__(self):
    Sets up ros stuff
    # Initialization
    self.interface = ObjectInterface()
    self.cvNet = cv.dnn.readNetFromCaffe("/home/kal3-5/workspaces/my_first_ws/src/tensorflow_ros_tool/src/tensorflow_ros_tool/models/caffe/MobileNetSSD_deploy.prototxt","/home/kal3-5/workspaces/my_first_ws/src/tensorflow_ros_tool/src/tensorflow_ros_tool/models/caffe/MobileNetSSD_deploy.caffemodel")
    # Register callbacks for subscriber and dynamic reconfigure
    self.reconfigure = Server(ObjectConfig, self.reconfigure_callback)
    self.interface.dummy_subscriber.registerCallback ...
edit retag flag offensive close merge delete

2 Answers

Sort by » oldest newest most voted

answered 2018-05-16 13:12:12 -0600

There are issues with getting the most recent message using the message passing system. If you're just working with bag files there may be a better option all together. Have you considered using the rosbag API?

This would allow you playback the bag in your own code and always extract the most recent message.

edit flag offensive delete link more


The problem is that I have already stopped using the rosbag, and began using a real environment, and I still get the delay, but not that big, maybe 2 seconds

erivera1802 gravatar image erivera1802  ( 2018-05-17 08:26:34 -0600 )edit

Okay, this might be overkill but you could make your own camera driver using OpenCV and connect to it using a service instead of a topic. This service could always return the most recent image captured.

PeteBlackerThe3rd gravatar image PeteBlackerThe3rd  ( 2018-05-17 09:09:13 -0600 )edit

answered 2018-05-17 07:38:08 -0600

updated 2018-05-18 04:10:08 -0600

This sounds like you use a too big queue size for your image subscription. If your image callback is faster than the rate at which subscribed images come in, things are fine.

If the callback takes longer, the next image in the queue is processed while any new incoming images are added to the end of the queue. You thus get huge latency in your processed images. The quick fix is dropping the queue size to 1. In that case, old images are dropped. Another option is decoupling callbacks from processing and performing both in different threads.

/edit: Thanks for the hints in the comments and apologies for missing that queue size was already set to 1. So it appears for rospy, it actually has to be set to queue_size=None , see Q/A 269855.

edit flag offensive delete link more


The OP has already stated they've used a queue size of 1.

PeteBlackerThe3rd gravatar image PeteBlackerThe3rd  ( 2018-05-17 08:06:39 -0600 )edit

Yes, but that is not how rospy works. Look for older Q&As about that here on ROS Answers.

gvdhoorn gravatar image gvdhoorn  ( 2018-05-17 08:10:53 -0600 )edit

Then what should I modify? Because yes, my queue size is already 1

erivera1802 gravatar image erivera1802  ( 2018-05-17 08:27:02 -0600 )edit

Updated answer and added link to the discussion @gvdhoorn referred to. Seems rospy is weird indeed ;)

Stefan Kohlbrecher gravatar image Stefan Kohlbrecher  ( 2018-05-18 04:11:33 -0600 )edit

Question Tools

1 follower


Asked: 2018-05-16 12:17:28 -0600

Seen: 1,767 times

Last updated: May 18 '18