Jinyan Hu

Engineering practice is the key bridge between college education and actual work in the industry. In order to deliver qualified talents with engineering quality to the industry, this paper explores integrating software engineering thinking into the Embedded System Design course. A practical and effective teaching mode is designed consisting of immersive learning, case-based learning, progressive practice, interactive learning, and autonomous learning. Through this teaching mode, multi-levels of closed-loop have been established including final project cycle closed-loop, testing cycle closed-loop, and product cycle closed-loop. During this process, students gradually transition to putting forward product requirements, carrying out design and development, thinking and solving problems, collaborating, and assuring quality from the perspective of software engineering. The practice results show that students’ engineering quality has been significantly improved.

High-accuracy detection of vacant parking slots using visual system is a prerequisite for automatic parking, which is significant for the promotion of automatic parking. Complex visual environment such as light occlusion, intensity variations and visual accuracy raises significant challenges for vision-based system parking slot detection. To address such problems, REPSDet (Resnet50 based Parking Slot Detection) is proposed for parking slot detection and occupancy classification in this paper, which is based on deep learning. Parking slot proximity information could be fully extracted by REPSDet,which is combined with parking slot classification and location information to improve the parking slot detection effect. For parking slot detection, REPS-Net is proposed in this paper, which is an improved yolov4 parking slot detection method. In REPS-Net, marker point positioning combined with geometric clues are used to infer the precise location of parking slots. For occupancy classification, the deep convolutional network m\_DCNN is proposed in this paper. Network structure and operation performance are optimized by m_DCNN, and the accuracy of parking slot detection is further improved. Experimental results show that REPSDet achieves precision rate of 97.77% and recall rate of 99.38% on the PS2.0_A dataset, which is significantly better than compared methods.

With the application of the random sampling method in the down-sampling of point clouds data, the processing speed of point clouds has been greatly improved. However, the utilization of semantic information is still insufficient. To address this problem, we propose a point cloud semantic segmentation network called MFFRand (Multi-Scale Feature Fusion Based on RandLA-Net). Based on RandLA-Net, a multi-scale feature fusion module is developed, which is stacked by encoder-decoders with different depths. The feature maps extracted by the multi-scale feature fusion module are continuously concatenated and fused. Furthermore, for the network to be trained better, the multi-loss supervision module is proposed, which could strengthen the control of the training process of the local structure by adding sub-losses in the end of different decoder structures. Moreover, the trained MFFRand network could be connected to the inference network by different decoder terminals separately, which could achieve the inference of different depths of the network. Compared to RandLA-Net, MFFRand has improved mIoU on both S3DIS and Semantic3D datasets, reaching 71.1% and 74.8%, respectively. Extensive experimental results on the point cloud dataset demonstrate the effectiveness of our method.