KCF Object Tracking (No-PTZ)

1. Experiment Objective

The KCF (Kernelized Correlation Filter) tracking algorithm is used to track any object manually selected with a bounding box, and the car rotates to track so that the target stays at the center of the frame.

2. Experiment Principle

1. Keyword Explanation

2. Technical Architecture

/camera/image_raw (BEST_EFFORT)
  → kcf_tracker_node
    ├── Initialization: select an ROI with the mouse → cv.TrackerKCF_create().init()
    ├── Tracking loop: tracker.update() → new tracking box → centroid calculation
    └── Body control timer (10Hz): pixel error → PD angular velocity → /cmd_vel (BEST_EFFORT)

Note: This feature only supports track (tracking) mode; follow and control modes are not supported. Once the KCF tracker loses the target, it cannot recover automatically.

3. Core Topic Quick Reference

Firmware Downlink

3. Experiment Steps

1. Hardware Preparation

2. Start the Agent + Camera + Joint Model (Terminal 1)

xxxxxxxxxx
ros2 launch microros_v2_bringup car_base.launch.py

3. Start KCF Object Tracking (Terminal 2)

xxxxxxxxxx
ros2 launch microros_v2_no_ptz_visual_control_pkg kcf_tracker.launch.py

5. Operation Instructions

4. Experiment Phenomenon

• After selecting a target, the tracking box follows the target and the car automatically rotates to keep the target at the center of the frame
• After the target moves out of the frame or is fully occluded, the frame shows "Target Lost!"
• The KCF tracker cannot recover automatically after the target reappears; you need to manually re-select the target
• The ESP32 camera has a low frame rate (about 7 fps); it is recommended to select a larger, feature-rich target and move it slowly

5. Code Explanation

Source path:

• Launch: ~/microros_ws/src/microros_v2_no_ptz_visual_control_pkg/launch/kcf_tracker.launch.py
• Node source: ~/microros_ws/src/microros_v2_no_ptz_visual_control_pkg/microros_v2_no_ptz_visual_control_pkg/kcf_tracker_node.py

1. Core Node Analysis

kcf_tracker_node uses the OpenCV Tracker API. Core workflow:

• Initialization: waits for the user to select an ROI with the mouse → press Space → create cv.TrackerKCF_create() (or MIL/CSRT) → tracker.init(frame, bbox)
• Tracking loop (15Hz): each frame ok, bbox = tracker.update(frame) → on success, update the tracking box position + compute the centroid deviation → PD outputs angular velocity
• Loss detection: ok=False or bbox out of the frame → mark as lost (Lost), wait for re-selection
• KCF limitation: no automatic re-detection after loss; the user must manually re-initialize

2. Key Parameters

Parameter definition file:

• Launch: ~/microros_ws/src/microros_v2_no_ptz_visual_control_pkg/launch/kcf_tracker.launch.py
• Node: ~/microros_ws/src/microros_v2_no_ptz_visual_control_pkg/microros_v2_no_ptz_visual_control_pkg/kcf_tracker_node.py

3. Node Description

6. Common Problems