Regular inspection of sewer pipes can detect serious defects in time, which is significant to ensure the healthy operation of sewer systems and urban safety. Currently, the widely used closed-circuit television (CCTV) inspection system relies mainly on manual assessment, which is labor intensive and inefficient. Therefore, it is urgent to develop an efficient and accurate automatic defect detection method. In this paper, an improved method based on YOLOv4 is proposed for the detection of sewer defects. A significant improvement of this method is using the spatial pyramid pooling (SPP) module to expand the receptive field and improve the ability of the model to fuse context features in different receptive fields. Meanwhile, the influence of three bounding box loss functions on model performance are compared based on their processing speed and detection accuracy, and the effectiveness of the combination of DIoU loss function and SPP module is verified. In addition, to address the lack of datasets for sewer defect detection, a dataset that contains 2700 images and 4 types of defects was created, which provides useful help for the application of computer vision techniques in this field. Experimental results show that, compared with the YOLOv4 model, the mean average precision (mAP) of the improved model for sewer defect detection are improved by 4.6%, the mAP can reach 92.3% and the recall can reach 89.0%. The improved model can effectively improve the detection and classification accuracy of sewer defects, and has significant advantages compared with other methods.

To address the issues of low automation, reliance on manual screening by professionals, and long detection cycles in current urban drainage pipeline defect detection, this study proposes an improved object detection algorithm called EFE-SSD (enhanced feature extraction SSD), based on the SSD (Single Shot MultiBox Detector) network. Firstly, the RFB_s module is added to the SSD backbone network to enhance its feature extraction capabilities. Additionally, multiple scale features are fused to improve the detection performance of small target defects to some extent. Then, the improved ECA attention mechanism is used to adjust the channel weights of the output layer, suppressing irrelevant features. Finally, the Focal Loss is employed to replace the cross-entropy loss in the SSD network, effectively addressing the issue of imbalanced positive and negative samples during training. This increases the weight of difficult-to-classify samples during network training, further improving the detection accuracy of the network. Experimental results show that EFE-SSD achieves a detection mAP of 92.2% for four types of pipeline defects: Settled deposits, Displaced joints, Deformations, and Roots. Compared to the SSD algorithm, the model’s mAP was increased by 2.26 percentage points—ensuring the accuracy of pipeline defect detection.

The study of the intricate flight patterns and behaviors of swarm insects, such as drosophilas, has long been a subject of interest in both the biological and computational realms. Tracking drosophilas is an essential and indispensable method for researching drosophilas’ behaviors. Still, it remains a challenging task due to the highly dynamic nature of these drosophilas and their partial occlusion in multi-target environments. To address these challenges, particularly in environments where multiple targets (drosophilas) interact and overlap, we have developed a long-term Trajectory 3D Position and Orientation Tracking Method (Long 3D-POT) that combines deep learning with particle filtering. Our approach employs a detection model based on an improved Mask-RCNN to accurately detect the position and state of drosophilas from frames, even when they are partially occluded. Following detection, improved particle filtering is used to predict and update the motion of the drosophilas. To further enhance accuracy, we have introduced a prediction module based on the self-attention backbone that predicts the drosophila’s next state and updates the particles’ weights accordingly. Compared with previous methods by Ameni, Cheng, and Wang, our method has demonstrated a higher degree of accuracy and robustness in tracking the long-term trajectories of drosophilas, even those that are partially occluded. Specifically, Ameni employs the Interacting Multiple Model (IMM) combined with the Global Nearest Neighbor (GNN) assignment algorithm, primarily designed for tracking larger, more predictable targets like aircraft, which tends to perform poorly with small, fast-moving objects like drosophilas. The method by Cheng then integrates particle filtering with LSTM networks to predict particle weights, enhancing trajectory prediction under kinetic uncertainties. Wang’s approach builds on Cheng’s by incorporating an estimation of the orientation of drosophilas in order to refine tracking further. Compared with those methods, our method performs with higher accuracy on detection, which increases by more than 10% on the F1 Score, and tracks more long-term trajectories, showing stability.

Accident detection using Closed Circuit Television (CCTV) footage is one of the most imperative features for enhancing transport safety and efficient traffic control. To this end, this research addresses the issues of supervised monitoring and data deficiency in accident detection systems by adapting excellent deep learning technologies. The motivation arises from rising statistics in the number of car accidents worldwide; this calls for innovation and the establishment of a smart, efficient and automated way of identifying accidents and calling for help to save lives. Addressing the problem of the scarcity of data, the presented framework joins Generative Adversarial Networks (GANs) for synthesizing data and Convolutional Neural Networks (CNN) for model training. Video frames for accidents and non-accidents are collected from YouTube videos, and we perform resizing, image enhancement and image normalisation pixel range adjustments. Three models are used: CNN, Fine-tuned Convolutional Neural Network (FTCNN) and Vision Transformer (VIT) worked best for detecting accidents from CCTV, obtaining an accuracy rate of 94% and 95%, while the CNN model obtained 88%. Such results show that the proposed framework suits traffic safety applications due to its high real-time accident detection capabilities and broad-scale applicability. This work lays the foundation for intelligent surveillance systems in the future for real-time traffic monitoring, smart city framework, and integration of intelligent surveillance systems into emergency management systems.