Edge Detection Task

Tasks and Results

1. Basic

For C++, the EdgeDetector.hpp and EdgeDetector.cpp files should be implemented. Any required libraries should be added to CMakeLists.txt. For Python, edge_detector.py should be completed with installation instructions for dependencies.

It should be possible to provide different images as input and display the output with edges detected in green. Additionally, a README.md file should include installation instructions, implementation steps, concepts used, and potential improvements.

Result

Edge detection identifies rapid changes in pixel intensity. The implemented solutions are:

1. Laplacian Edge Detector: Identifies edges as zero-crossing points in the second derivative (2D Laplacian) of the image function.
2. Canny Edge Detector: Uses a gradient operator (Sobel filter) to compute gradients at each pixel, then applies a 1D Laplacian along the gradient orientation. Strong zero crossings are classified as edges.

Implementation Details:

• The image is converted from RGB to grayscale to simplify computation.
• A Gaussian blur is applied to reduce noise, preventing false zero crossings.
• The function edge_detect in edge_detector.py implements these methods:
  • 0: Canny Edge Detector (default)
  • 1: Laplacian Edge Detector
• A commented-out section overlays results from both detectors for comparison.

2. Vision_ROS

Task:

Provide ROS .srv and .msg files to create a ROS service for edge detection. Implement a client that detects edges in images from a directory.

Result:

• A ROS service named Image_edge_detect_service accepts input_img_path (string) and returns output_img_path (string), storing the edge-detected image.
• To use:
  • Run ../src/edge_detect_img_node.py to start the service.

Additional Task: Detect edges in images from a ROS topic when playing a .bag file, then visualize input and output images in RViz.

Result:

• A ROS node image_raw_listener subscribes to /camera/color/image_raw to receive sensor_msgs/Image, processes edge detection, and republishes results to edge_detected_image.
• To perform this:
  1. Play a provided .bag file: rosbag play --clock -l <path_to_bagfile>
  2. Run ../src/edge_detect_bag_data_node.py
  3. Add RViz displays for /camera/color/image_raw and edge_detected_image

Additional Task: Convert detected edge pixels from (u, v) to 3D (x, y, z) using depth images and camera parameters. Publish 3D data to a ROS topic (edge_points) of type sensor_msgs/PointCloud.

Result:

• Using intrinsic camera parameters (/camera/color/camera_info) and depth data (/camera/depth/image_rect_raw), the 3D coordinates are computed as:
  • x = depth * (u - cx) / fx
  • y = depth * (v - cy) / fy
  • z = depth
• The edge_points topic publishes sensor_msgs/PointCloud.
• To perform this:
  1. Play a .bag file: rosbag play --clock -l <path_to_bagfile>
  2. Run ../src/edge_detect_bag_data_node.py
  3. Start publishing edge_points: ../src/edge_pointcloud_pub_node.py
  4. Verify output: rostopic echo /edge_points

3. Robot_ROS

Task:

Visualize 3D edge points in RViz as markers alongside a robot model.

Result:

• Markers are published in camera_color_optical_frame since /camera/... topics use this frame.
• The edge_3D_marker topic publishes visualization_msgs/Marker.
• To view markers:
  1. Play a .bag file: rosbag play --clock -l <path_to_bagfile>
  2. Run ../src/edge_detect_bag_data_node.py
  3. Start point cloud publishing: ../src/edge_pointcloud_pub_node.py
  4. Open RViz and add a Marker display for /edge_3D_marker

Transforming 3D Data to Robot Frames:

• TransformListener.transformPoint() converts 3D coordinates from camera_color_optical_frame to robot frames using transformations from /tf and /tf_static.
• The transformation logic is implemented but commented out in ../src/edge_pointcloud_pub_node.py due to testing limitations.

Room for Improvement:

• Refactoring Python scripts into classes for better modularity.
• Translating rospy implementations into roscpp for efficiency.
• Enhancing noise filtering to improve edge detection accuracy.

---

This repository provides robust edge detection solutions with ROS integration, visualization in RViz, and 3D coordinate transformation for robotic applications.