Xingqiang Li

Unsupervised real-time stereo matching is of great research value in robot navigation due to its independence from ground truth and real-time efficiency. The core challenge lies in the design of loss functions that can provide accurate guidance and the efficient network architectures. The commonly used photometric loss is prone to provide incorrect guidance because of the influence of reflection, left-right color inconsistency, low texture, and occlusion. As a weak supplement, the smoothness loss can ameliorate the multi-solution problems caused by low texture, but it is not effective for strong incorrect guidance caused by the other problems. In order to provide more accurate and powerful supplementary guidance, a 3D planar segmentation loss is proposed with advancements in addressing the strong incorrect guidance problem, which could be generally integrated into traditional unsupervised training losses. Furthermore, the real-time stereo matching approach of the hierarchical iterative tile refinement network is applied to unsupervised stereo matching, with necessary modifications to address the detrimental architectures that hinder its performance in unsupervised training. Experimental results verify the effectiveness of the 3D planar segmentation loss and the network modification. The proposed pipeline achieves competitive accuracy compared to existing unsupervised stereo matching methods while maintaining real-time efficiency.

Cylindricity is a crucial factor to estimate the manufacturing quality of large rotary parts. However, existing measurement systems can hardly deal with the precise cylindricity measurement challenges posed by large dimensions and great weights. In this article, a novel measurement system along with an improved error separation model and efficient solution algorithm for large rotary parts is proposed. Concretely, the innovative research includes three parts. First, a precise cylindricity measurement system with large variable ranges is proposed. Different from traditional workpiece rotation means, a sensor rotation system for data acquisition is designed to be independent of the air-bearing platform for workpiece centering. This contributes to stable and precise measurement under high-load and field conditions. Second, a more sophisticated and accurate measurement model is established, and the model-based error separation method is presented, which enables more accurate separation of error sources, such as eccentricity, measuring line offset, and workpiece axis tilt. Finally, a differential evolution growth algorithm (DEG), which is able to balance global search ability and local information utilization, is put forward to improve the efficiency and accuracy of model fitting in the error separation process. The proposed measurement system, model, and algorithm are all verified by simulations and experiments.

In the process of metal surface defect detection, it is difficult to detect and segment small defects. In order to solve this problem, this paper uses weighted object variance algorithm to detect metal surface defects. Then the feature of defect area is extracted and the defect classification model based on support vector machine is trained. In order to verify the effectiveness of the method, this paper takes the metal surface of bearing cylindrical roller as the specific research object. The experimental results show that the method meets the production requirements.

For online quality control of car engine production, pneumatic measurement instrument plays an unshakeable role in measuring diameters inside engine block because of its portability and high-accuracy. To the limitation of its measuring principle, however, the working space between the pneumatic device and measured surface is too small to require manual operation. This lowers the measuring efficiency and becomes an obstacle to perform automatic measurement. In this article, a high-speed, automatic measuring system is proposed to take the place of pneumatic devices by using a laser-based measuring unit. The measuring unit is considered as a set of several measuring modules, where each of them acts like a single bore gauge and is made of four laser triangulation sensors (LTSs), which are installed on different positions and in opposite directions. The spatial relationship among these LTSs was calibrated before measurements. Sampling points from measured shaft holes can be collected by the measuring unit. A unified mathematical model was established for both calibration and measurement. Based on the established model, the relative pose between the measuring unit and measured workpiece does not impact the measuring accuracy. This frees the measuring unit from accurate positioning or adjustment, and makes it possible to realize fast and automatic measurement. The proposed system and method were finally validated by experiments.