With the rapid development of autonomous driving technology, accurate perception of the surrounding environment has become increasingly critical, and 3D environment perception has emerged as a major research focus in this field. Traditional 3D perception systems rely heavily on expensive sensors such as LiDAR, which offer high accuracy but incur substantial costs and computational demands, limiting their scalability in large autonomous vehicle fleets. Although more recent 3D occupancy prediction methods rely solely on multicamera inputs, they typically require supervised learning with annotated 3D occupancy data, which is costly to obtain and consumes substantial memory. To address these challenges, this article proposes Image2Occupancy, an improved 3D Gaussian-splatting-based occupancy prediction method that uses only 2D surround-view camera images. The method enables effective semantic occupancy prediction of 3D scenes while reducing the need for annotated data and large memory capacity.

The Image2Occupancy framework consists of two components: (1) 2D-to-3D feature extraction and spatial mapping, and (2) self-supervised 3D occupancy representation learning. In the first component, BEVStereo and Swin Transformer modules extract 2D features from panoramic input images. These features are then interpolated and mapped to 3D space using the intrinsic and extrinsic parameters of the camera, yielding voxel-level feature representations. This process converts 2D image information into 3D semantic occupancy cues, providing accurate input for subsequent self-supervised learning. In the second component, an improved Gaussian splatting technique projects 3D voxel features back onto the 2D image plane while preserving semantic information. Gaussian points placed at each voxel center approximate scene occupancy, enabling rendering of semantic and depth maps by computing pixel-level depth and semantic information. A novel self-supervised learning framework generates pseudo-labels from the predicted depth and semantic maps of the model, eliminating the need for real 3D occupancy labels. A specialized loss function, combining cross-entropy and depth losses, minimizes discrepancies between rendered and ground-truth semantic and depth maps, optimizing prediction accuracy.

Experiments on the NuScenes dataset show that Image2Occupancy achieves an mIoU of 27.87, improving performance by 3.94 percentage points (a 16.5% increase) over existing 2D-input methods and performing comparable to, or better than, several 3D-input methods. Compared with NeRF-based approaches, GPU memory usage is reduced by 54.7% while maintaining the same number of Gaussian points. Ablation studies further validate the effectiveness of the core components of the method.

Image2Occupancy reduces hardware dependence and substantially decreases the need for large annotated datasets through self-supervised learning, offering a cost-effective and scalable 3D environment perception solution for autonomous driving systems with strong potential for practical deployment.

Due to the difficulties in experimental conditions and implementation, traditional first-order RC transition processes were only designed to verify theoretical knowledge. Many engineering elements that must be carefully considered in practical circuit design and debugging cannot be excavated and reflected. This may lead to a weak engineering background of experiment content, limited exploration space for students, and insufficient inspiration for students, which is not conducive to the improvement of the “two-property and one-degree” of the engineering course. To address these issues, this paper has made significant reforms in the experimental method, content, and teaching methods.

First, the experiment method has been reformed using a DC stabilized power supply as a circuit excitation instead of square waves, similar to traditional experiments. The circuit is controlled by a mechanical switch to start the charging and discharging process of the capacitor. The instantaneous charging and discharging waveforms of the capacitor are captured through the single trigger function of the digital storage oscilloscope, exposing and displaying multiple “abnormal” phenomena that seem to be inconsistent with theoretical analysis, which facilitates students' active thinking and improves the innovation of the experiment. Second, the experiment content has been upgraded using heuristic teaching methods to guide students to deeply explore the “abnormal” phenomena occurring during the experimental process. We then study the engineering characteristics widely existing in actual circuits, including stray capacitance, switch jitter, and instrument input impedance. Theoretical knowledge will be used to analyze and explain these actual phenomena and characteristics, and improvement and avoidance plans will be proposed to promote students' intuitive understanding of actual circuits and improve the high order of the experiment. Finally, the teaching methods have been updated, and a hybrid intelligent learning loop consisting of online SPOC learning and offline intelligent experiments has been constructed. The “Circuit Experimental Technology” SPOC course has been developed on the MOOC website of Chinese universities, with over 30 course preview and operation explanation videos uploaded. The students preview the course before class through the SPOC course, laying a strong foundation for their knowledge. Afterward, they come to the laboratory and conduct self-directed learning of first-order RC circuit experiments using the intelligent teaching system developed by our team, following the new experimental process of “six steps of self-directed learning,” which includes “automatic preview test” “self-directed video learning” “intelligent experiment operation” “intelligent assisted troubleshooting” “whole process accompanying evaluation” and “electronic report submission”. After completing the experimental operations, students will continue to review knowledge and take unit tests on the SPOC platform after class to complete this experimental learning. Based on the aforementioned measures, students have been promoted to independently learn the assessment of the experiment process, and the challenges associated with the experiment have increased.

This experiment project has been normally used in the “Circuit Experimental Technology” course of our school for five years, with a total of 130 classes and more than 4000 students benefiting from this course. Based on a series of educational reform measures with this project, the “Circuit Experimental Technology” was awarded the title of “High Quality Undergraduate Course in Beijing Universities” in 2022, and the teaching staff was also awarded the title of “Excellent Professional Course Lecturer in Beijing Universities”.

Recent practice has shown that the reformed first-order RC circuit experiment helps deepen students' understanding of the essence and connotation of the dynamic circuit response process, improves their ability to analyze problems and debug circuits, exercises their thinking, and enhances engineering accomplishments. The outcomes indicate the high application potential of the experiment.