ApexVision

Inspiration

Every racer, whether on a real circuit or simulator, dreams of mastering the perfect line — the path that balances speed, control, and tire life. But analyzing laps manually or through telemetry takes hours. We wanted to build an AI that sees what the driver feels — detecting friction, drift, and wear visually — and then teaches the most efficient way to drive the track.

What it does

ApexVision is an AI-powered Visual Difference Engine for racing optimization. It analyzes onboard or trackside videos of multiple laps along with a top-view circuit image to:

Detect high-friction, drift, and tire-wear zones.

Reconstruct a 3D view of the track.

Predict the most efficient racing line that minimizes tire wear and maintains speed.

Visualize the “optimal lap” with heatmaps and 3D overlays — like a Google Maps for racing lines.

How we are going to build it

Computer Vision: Using OpenCV for frame differencing and optical flow to detect motion and friction zones.

Visual Analysis Engine: Aggregated multiple laps to generate heatmaps showing high-stress track areas.

Optimization: Applying heuristic pathfinding (A* / Dijkstra) to find the lowest-cost racing line.

3D Visualization: Using Open3D and Matplotlib to render the reconstructed circuit and overlay predicted lines.

Interface: Streamlit prototype for live demo and interactive visualization.

Our aim during the Hackathon

To develop an AI-powered visual difference engine capable of analyzing racing footage lap-by-lap.

To detect and visualize tire wear, drift, and friction zones using computer vision techniques.

To reconstruct the track in 3D and overlay performance heatmaps for better visual interpretation.

To predict the optimal racing line that minimizes tire wear while maintaining maximum speed.

To demonstrate how visual intelligence can transform racing analytics into actionable insights.

To build a functional and visually engaging prototype that showcases the potential of AI in motorsport sustainability and performance optimization.

What we learned

The power of combining computer vision with domain-specific reasoning (racing physics).

How small visual cues (like skid marks or lateral movement) can reveal deep performance insights.

Rapid prototyping under pressure — balancing model accuracy with demo readiness.

What's next for ApexVision

Integrate real telemetry (speed, throttle, G-force) for hybrid AI optimization.

Build a driver feedback system — personalized coaching using visual and performance data.

Launch ApexVision Cloud, allowing sim racers and teams to upload laps, compare lines, and receive instant analysis.

Built With

• github
• kaggle
• matplotlib
• numpy
• open3d
• opencv
• python
• scikit-learn
• streamlit