Tire appearance defect detection based on machine vision is an effective technology to improve the tire production quality. The detection process can be completed by the way of non-destructive testing. Therefore, more and more researchers are paying attention to this technology. However, tires are characterized by single block colors and various defects. It is a great challenge to accurately detect tire appearance defects. To complete the task of detecting tire defects, this paper presents a novel tire appearance defect detection method via combining histogram of oriented gradients (HOG) and local binary pattern (LBP) features. First, we construct a tire image dataset to provide defective and normal tire images. Then, histogram of oriented gradients and local binary pattern features of tire images are, respectively, extracted and used to train the support vector machine (SVM) classifier. Finally, the support vector machine classifier calculates the prediction scores of the test images via combining the histogram of oriented gradients and local binary pattern features. These scores can be utilized to determine whether the test image is a defective or a normal tire image, and the goal of tire appearance defect detection is achieved. Conducted on the tire image dataset, our method has verified the effectiveness of detecting tire detects, and the mean accuracy is improved more than 1.6% than the algorithm that only uses the histogram of oriented gradients or local binary pattern feature. The experimental results demonstrate that the combination of HOG and LBP features can increase tire appearance defect detection accuracy.

Multi-object tracking aims to track multiple objects across consecutive frames in a video, assigning a unique classifier to each object. However, issues such as occlusions, directional changes, or shape alterations can cause appearance variations, leading to detection and matching problems that in turn result in frequent ID switches. To solve these issues, this paper proposes a two-stage multi-object tracking framework based on a spatial and temporal fusion algorithm. First, the video frames are processed by a detector to identify objects and form rectangular detection areas. Meanwhile, an estimator predicts the target rectangular areas in the next frame. Then, we extract the optical flow of the target pixels within the detection and prediction areas, and then a temporal information model is established by calculating the average of the target pixels’ optical flow. Afterward, we present a spatial information model using the R-IoU (Reverse of Intersection over Union) between the detection and prediction areas. This spatial and temporal information is combined with weighted matrix fusion, which achieves the feature matching and association task. Finally, we implement a two-stage association multi-object tracking model using the mentioned fusion algorithm. Experiments on the MOTChallenge dataset using the official detector show that our two-stage multi-object tracking method based on the spatial and temporal fusion algorithm is robust in handling occlusions and ID switch issues. As of the submission of this paper, the proposed method has achieved the top ranking in the MOT17 benchmark when evaluated with the official detector.

Fast and accurate indoor occupancy detection is critical for energy efficiency and emergency rescue in the fields of smart building and indoor positioning. However, existing image-based indoor occupancy detection models often neglect small human targets and suffer from large parameters, compromising detection accuracy, real-time performance, and deployment on resource-constrained devices. To address these issues, this study proposes a modified lightweight indoor occupancy detection model based on YOLO v8. Firstly, a patch expanding layer is added to the neck of the YOLO v8 model for reshaping the feature maps of adjacent dimensions into higher-resolution feature maps. Secondly, the standard convolution in the original neck is replaced with the GSConv, boosting the non-linear representation by adding DSC layers and a shuffle operation, efficiently preserving hidden connections between channels. Additionally, the VoV-GSCSP in the neck is designed to adopt one-shot aggregation with GS bottlenecks based on GSConv, followed by a cross-stage partial network module. Experiments on the SCUT-HEAD dataset show the modified lightweight YOLO v8 reduces parameters by 9.3% and computational complexity by 8.6%, while increasing mAP50 by 1.4% compared to the baseline. The proposed model can detect indoor occupancy in a fast and precise manner.