As a key equipment for production and transportation, conveyor belts are widely used in the coal mining industry. Once the conveyor belt longitudinal tear occurs, it will seriously affect the production and even cause personal injury. Therefore, the longitudinal tear detection of the conveyor belt is extremely important. In this paper, the sound detection is introduced into the longitudinal tear detection of the conveyor belt, and an audio-visual detection method for conveyor belt longitudinal tear is proposed. Camera and microphone array are used to collect the image and sound signals of conveyor belt, and the conveyor belt is detected from both image and sound, respectively. Then the image and sound analysis results are combined to comprehensively judge the status of the conveyor belt. The experimental results show that the audio-visual detection method can accurately identify the normal, abnormal, and longitudinal tear of the conveyor belt. The detection accuracy is over 86.72% and the sensitivity of longitudinal tear detection is greater than 92.59%. The proposed audio-visual detection method is verified to meet the requirements of longitudinal tear detection of coal mining industry conveyor belts.

Three-dimensional information acquisition is crucial for the intelligent control and safe operation of bulk material transportation systems. However, existing visual measurement methods face challenges, including difficult stereo matching due to indistinct surface features, error accumulation in multi-camera calibration, and unreliable depth information fusion. This paper proposes a three-dimensional reconstruction method based on multi-camera hierarchical calibration. The method establishes a measurement framework centered on a core camera, enhances material surface features through speckle structured light projection, and implements a ‘monocular-binocular-multi-camera association’ calibration strategy with global optimization to reduce error accumulation. Additionally, a depth information fusion algorithm based on multi-epipolar geometric constraints improves reconstruction completeness through multi-view information integration. Experimental results demonstrate excellent precision with absolute errors within 1 mm for features as small as 15 mm and relative errors between 0.02% and 2.54%. Compared with existing methods, the proposed approach shows advantages in point cloud completeness, reconstruction accuracy, and environmental adaptability, providing reliable technical support for intelligent monitoring of bulk material transportation systems.