Skip to content

🧠 MSight Algorithms Overview

🚀 A Full Stack of Roadside Intelligence

MSight includes a powerful suite of built-in algorithms that enable perception, prediction, fusion, and safety analytics directly from roadside sensors. These algorithms are modular, high-performance, and designed to run efficiently at the edge, in the cloud, or in hybrid deployments.

MSight’s algorithm stack currently includes:

  • 2D object detection on fisheye/regular cameras
  • 3D object detection with LiDAR and camera fusion
  • Multi-sensor fusion for unified world-frame perception
  • Trajectory prediction using motion forecasting models
  • Near-miss and crash risk detection

Together, they provide a comprehensive, end-to-end perception and safety pipeline suitable for research, deployment, and production-scale intelligent infrastructure.

🎥 2D Object Detection

MSight supports real-time 2D detection using fisheye and standard cameras, using models such as YOLOv12 or any custom lightweight camera models optimized for roadside scenes.

⟶ Learn more: 2D Detection

📡 3D Detection

Using LiDAR, MSight provides detection algorithm that draws 3D bounding boxes using BEV-based detection networks.

⟶ Learn more: 3D Detection

🔄 Multi-Sensor Fusion

MSight includes Bayessian fusion strategies that fusion sensor data on Trajectory-level. This allow multi-camera + LiDAR deployments to produce a single world-aligned perception output.

Tip

Earlier-stage fusion methods (such as point cloud merging) are typically handled inside the detection pipeline itself. Because of this design, MSight focuses on providing late fusion at the trajectory level, which is more flexible and sensor-agnostic.

⟶ Learn more: Advanced Sensor Fusion

🎯 Multi-Object Tracking

MSight employs multi-object tracking (MOT) to maintain consistent object identities across frames. We provide Simple Online Realtime Tracking (SORT).

⟶ Learn more: Tracking Algorithms (coming soon)

🛣️ Trajectory Prediction

MSight integrates state-of-the-art motion forecasting models to estimate future trajectories of vehicles and pedestrians. Features include:

  • Multi-modal predictions (multiple possible futures)
  • Lane-aware motion prediction
  • Gaussian mixture modeling for future distribution

⟶ Learn more: Prediction Algorithms

⚠️ Near-Miss Detection

The system provides real-time safety analytics, detecting Vehicle–vehicle and vehicle–VRU conflicts.

Algorithms can be seamlessly integrated into the MSight system to form a complete safety pipeline—for example, triggering V2X warnings or uploading safety events to the cloud.

⟶ Learn more: Near-Miss Detection

📚 Continue Exploring