A group of researchers from the Hefei Institutes of Physical Science of the Chinese Academy of Sciences have proposed SCNet3D, a new 3D object detection system, to better identify surrounding objects. This system utilizes information from cloud data to improve small target detection.
The LiDAR-based 3D object detection system is crucial for safety and efficiency in autonomous cars. The traditional detection methods usually convert unordered cloud data into pseudo-images to fetch ordered information.
However, this conversion sometimes leads to limited feature extraction capabilities and tends to lose key information. This eventually leads to inferior detection accuracy, especially for small objects. In this context, the feature is structured information extracted from the point cloud data.
A new study published in IEEE Transactions on Intelligent Transportation Systems puts forth a 3D object detection method based on data augmentation and attention mechanisms. SCNet3D focuses on addressing features and data to enhance preserving data to catch the smaller objects.
“We propose SCNet3D, a novel pillar-based method that tackles the challenges of feature enhancement, information preservation, and small target detection from the perspectives of features and data,” mentions the study.
“It can help self-driving cars better detect small objects,” said the lead author Wang Zhiling.
The SCNet3D method embeds Feature Enhancement Module (FEM), which uses the attention mechanism to enhance 3D features layer by layer. For this, FEM utilizes important features across three dimensions from across the globe.
Self-driving AI wheelchair to aiding people with disabilities
Followed by FEM is the STMod-Convolution Network (SCNet) design. SCNet has two channels for fetching the features. One channel works on basic features. Meanwhile, the other channel manages complex, advanced features. This design combines the Birds Eye View (BEV) pseudo-images through two channels to achieve sufficient features.
Additionally, the Shape and Distance Aware Data Augmentation (SDAA) approach is used to add more sample points of the feature.
The experiments with SCNet3D have proven useful in detecting nearby objects, even in an environment with plenty of interference. Since AI is taking over roads via autonomous cars, this system makes a promising tool for self-driving cars.
Autonomous cars working together can speed up traffic by 35 percent
Journal Reference
- Yu Zhang, Lin Zhang, “Detection of Pavement Cracks by Deep Learning Models of Transformer and UNet”, IEEE Transactions on Intelligent Transportation Systems. DOI: 10.1109/TITS.2024.3486324