Enhancing Feature Tracking Reliability for Visual Navigation using Real-Time Safety Filter

🚀 Making Visual Navigation More Reliable

Enhancing Feature Tracking with a Real-Time Safety Filter

In environments without GPS, many robots rely on visual SLAM — using a camera to track their position by detecting and following visual features (like corners or edges). But here’s the catch:

🔍 If the number of visible features suddenly drops, SLAM performance can fail dramatically — causing tracking loss, localization errors, or even crashes.

Our work addresses this with a simple question:
Can we make a robot proactively protect its feature visibility — before it’s too late?

🎯 Key Idea: A Real-Time Safety Filter for Features

We propose a real-time safety filter that runs alongside the robot’s control system. Instead of blindly following a planned velocity, the filter slightly modifies the velocity to keep enough visual features in view.

✨ Think of it like:

A safety assistant that whispers to the robot:
“Maybe slow down a bit so you don’t lose track of those corners.”

🔧 How It Works (In Simple Terms)

Here’s what happens at each time step:

1. 🧭 A controller gives a reference velocity (v_ref) to the robot.
2. 👁 The camera sees visual features and computes an information score — a measure of how “rich” the current features are.
3. 🧮 A quadratic program (QP) solves for a new velocity (v_filtered) that:
   • Is close to v_ref
   • Ensures the information score stays above a threshold

📊 Suggested Graphic:

A diagram showing:

• Original velocity vector
• Modified velocity vector
• Feature score threshold
• Features moving out of FOV

🧪 Experiments

✅ Simulation:

• We tested in a simulated indoor environment with sudden changes in feature visibility.
• Without our filter, the robot frequently lost track of features.
• With our filter, it slowed or adjusted direction to maintain trackability.

Baseline Result

Proposed Result

Figure: Simulation Result

✅ Real-world Deployment:

• We mounted a monocular camera on a wheeled robot.
• The robot navigated safely even when entering textureless areas like blank walls or glass.

📷 Suggested Graphic:

Side-by-side screenshots of:

• Feature tracking over time (with and without filter)
• Trajectories diverging due to tracking failure

📈 Why It Matters

• 🎯 SLAM safety: Prevents catastrophic failures due to feature loss.
• ⚡ Real-time: The filter runs fast enough for real robots.
• 🧠 Minimal intervention: Only adjusts motion when necessary — keeps original behavior otherwise.

🧩 Future Ideas

We’re excited to:

• Combine this with active vision to move toward richer feature regions
• Extend it to multi-sensor fusion systems
• Use it in autonomous drones, where losing features mid-flight can be fatal

Bibtex

@article{kim2025enhancing,
  title={Enhancing Feature Tracking Reliability for Visual Navigation using Real-Time Safety Filter},
  author={Kim, Dabin and Jang, Inkyu and Han, Youngsoo and Hwang, Sunwoo and Kim, H Jin},
  journal={arXiv preprint arXiv:2502.01092},
  year={2025}
}