Range-Intensity-Profile-Guided Gated Light Ranging and Imaging Based on a Convolutional Neural Network.

Area of Science:

• Optics and Photonics
• Computer Vision
• Machine Learning

Background:

• Three-dimensional (3D) range-gated imaging provides high-resolution intensity and depth data.
• Existing depth recovery algorithms, like range-intensity correlation and deep learning, face challenges with data requirements or profile generation.
• Accurate depth information is crucial for various imaging applications.

Purpose of the Study:

• To develop a novel range-intensity-profile-guided method for accurate depth recovery from gated images.
• To overcome limitations of existing algorithms by leveraging synthetic data and a specialized neural network.
• To improve the performance of depth estimation in 3D range-gated imaging systems.

Main Methods:

• A convolutional neural network, termed range-intensity ratio and semantic network (RIRS-net), was developed.
• Synthetic training data was generated using Grand Theft Auto V, guided by the system's range-intensity profile (RIP).
• The RIRS-net was trained on synthetic data and fine-tuned with RIP data, learning both semantic and range-intensity depth cues.

Main Results:

• The proposed method demonstrated superior performance in depth recovery compared to existing algorithms.
• Evaluation experiments on both real-scene and synthetic datasets confirmed the method's effectiveness.
• The network successfully integrated semantic and range-intensity depth cues for enhanced accuracy.

Conclusions:

• The range-intensity-profile-guided approach offers a robust solution for depth recovery in 3D range-gated imaging.
• Utilizing synthetic data generation significantly reduces the need for extensive real-world training data.
• This method advances the capabilities of gated light ranging and imaging systems.