Search Results for « Road debris detection »

We present an algorithm to detect unseen road debris using a small set of synthetic models. Early detection of road debris is critical for safe autonomous or assisted driving, yet the development of a robust road debris detection model has not been widely discussed. There are two main challenges to building a road debris detector: first, data collection of road debris is challenging since hazardous objects on the road are rare to encounter in real driving scenarios; second, the variability of road debris is broad, ranging from a very small brick to a large fallen tree. To overcome these challenges, we propose a novel approach to few-shot learning of road debris that uses semantic augmentation and domain randomization to augment real road images with synthetic models. We constrain the problem domain to uncommon objects on the road and allow the deep neural network, HazardNet, to learn the semantic meaning of road debris to eventually detect unseen road debris. Our results demonstrate that HazardNet is able to accurately detect real road debris when only trained on synthetic objects in augmented images.

Road debris on roadways and airport runways poses significant hazards, thereby necessitating reliable monitoring. Various sensors, such as camera, LiDAR, and radar, have been employed for debris detection, among which a W-band synthetic aperture radar (SAR) system based on a 77 GHz frequency-modulated continuous wave radar has attracted increasing attention. Nevertheless, road debris with small size and low radar reflectivity is challenging to detect, as it is indistinguishable from the background in the SAR intensity image. In this study, we propose a road-debris-detection approach based on differential-phase contrast (DPC). DPC denotes the contrast in the differential phase between elevated debris and the surrounding ground surface. Since this contrast arises from height differences, DPC-based detection is less directly dependent on target-to-background intensity contrast than the conventional SAR intensity-based detection, thereby improving the separability of low radar cross-sectional (RCS) debris. A W-band SAR system for road-debris detection was constructed, and a DPC-based detection framework combining differential-phase-based detection with SAR intensity-based detection was proposed. The detection experiments were conducted on seven types of debris with diverse materials and shapes, including debris with low RCS. Compared with the conventional SAR intensity-based method, the proposed DPC-based detection framework achieved precision, recall, and F1-score of 77.78%, 100%, and 87.50%, respectively, representing improvements of 69.94, 42.86, and 73.71 percentage points. The debris-region differential phase exhibited an average contrast-to-noise ratio of 3.135, which was 2.829 higher than that of the intensity image, confirming that the height-induced DPC provides substantially stronger target-to-background contrast. This enhanced contrast enabled reliable detection of low-RCS road debris under the experimental conditions, demonstrating the effectiveness of height-induced DPC as a discriminative feature.

Highway road debris is often small, irregular, weakly textured, and poorly contrasted against complex backgrounds, while limited real-world samples constrain model generalization. This study constructs a Synthetic-Debris dataset by compositing three representative debris categories-brick, paper box, and rock-onto highway scenes to support small-object and low-contrast detection. FBENet (Frequency-Aware Bidirectional Feature Fusion and Efficient Attention Enhancement Network) is developed from YOLO11n using a task-oriented, stage-coupled design: FreqFusion (Frequency-aware Feature Fusion) is embedded at two top-down cross-scale fusion stages to preserve boundary details and improve cross-resolution consistency before learnable bidirectional aggregation by BiFPN (Bi-directional Feature Pyramid Network), while EMA (Efficient Multi-Scale Attention) is applied only to the final high-resolution P3 feature before detection. The contribution lies in the stage-specific organization of established operations rather than in proposing new primitive modules. On Synthetic-Debris, FBENet achieved an [email protected] of 0.862, an [email protected]:0.95 of 0.664, and an F1-score of 0.811, with 2.43 M parameters and 6.8 GFLOPs. In the fixed synthetic-to-real split, FBENet exceeded YOLO11n by 5.0 and 2.8 percentage points in [email protected] and [email protected]:0.95, respectively. Five-fold cross-validation showed modest mean AP gains and lower overall AP variance, although transfer behavior remained class-dependent. Runtime evaluation showed that FBENet incurred additional latency relative to YOLO11n. Under PyTorch FP32, the network inference latency increased from 8.07 to 12.79 ms, while the end-to-end throughput decreased from 79.15 to 53.97 FPS. Under TensorRT FP16, FBENet achieved 126.59 FPS compared with 137.64 FPS for YOLO11n. These results indicate that FBENet improves detection accuracy at the cost of additional runtime, representing an accuracy-latency trade-off rather than an improvement in inference efficiency. Overall, frequency-aware bidirectional fusion and selective high-resolution attention are useful for improving highway debris detection under limited real-data conditions.

Abstract. Earthquakes can cause considerable damage to transportation infrastructure, affecting emergency response. Timely detection of road debris is important in such situations. The purpose of this research is to address the challenges related to the rapid detection of collapsed areas and road blockages. Deep learning has become an essential tool in remote sensing and image processing, offering improved capabilities for classification, and segmentation from high-resolution imagery. Its integration into post-disaster analysis enables more accurate assessments compared to traditional methods. This study examines the performance of three commonly used semantic segmentation models Unet, Attention Unet (AttUnet), and ResUnet++ for road debris detection. Two earthquake events were used as case studies: the 2023 earthquake in Osmaniye, Türkiye (Mw 7.8), captured by the Pleiades satellite, and the 2017 earthquake in Sarpol-e Zahab, Kermanshah, Iran (Mw 7.3), captured by a Phantom 4 Pro drone. The models were evaluated using three metrics: Intersection over Union (IoU), Recall, and Accuracy. The results show that ResUnet++ achieved higher performance compared to Unet and AttUnet in both cases. For the Osmaniye dataset, ResUnet++ reached an IoU of 80.81%, Recall of 78.88%, and Accuracy of 96.12%. On the Sarpol-e Zahab dataset, it obtained an IoU of 81.62%, Recall of 80.28%, and Accuracy of 97.24%. Unet and AttUnet performed at lower levels across all evaluated metrics. This comparison provides a clear assessment of model performance in post-earthquake debris detection tasks and contributes to ongoing work in the application of deep learning and high-resolution imagery for geospatial analysis in disaster response contexts.

Abstract. Road debris poses a significant threat to traffic safety, particularly small-scale objects that are difficult to detect under complex road conditions. Traditional manual inspection methods suffer from low efficiency, limited coverage, and delayed response. To address these challenges, this study proposes a lightweight mobile monitoring system for small-scale road debris detection based on vehicle-mounted sensing and dynamic data transmission. The system integrates a lightweight image acquisition device, a self-developed mobile monitoring software, and an improved lightweight detection model termed Dynamic-YOLOv8n. The model introduces dynamic convolution and attention mechanisms to enhance feature representation while maintaining computational efficiency. In addition, a dynamic compression transmission mechanism is designed to adaptively adjust image compression and transmission frequency under fluctuating network bandwidth, ensuring stable data transfer between the front-end terminal and the backend platform.A dedicated debris dataset was constructed from field-collected images covering diverse road types and environmental conditions. Experimental results show that the proposed model achieves a mean Average Precision (mAP) of 93.2% while maintaining real-time detection performance. Field tests conducted on urban road networks further demonstrate that the system significantly improves detection efficiency compared with manual inspection and enables reliable real-time monitoring of road debris. The proposed framework provides an effective solution for intelligent road inspection and digital road maintenance.

Many automotive accidents are reported every year, some of which are attributed to fallen debris on road. For road debris detection, it would be difficult to find with imaging sensor while driving, especially in case of bad weather conditions such as rain and fog. The use of 79GHz-band ultra-wideband automotive radar is hence expected for road debris detection because of the high range-resolution and good weather resistance. The detection performance basically depends on the radar cross section (RCS) of road debris. For example, the maximum detection range is proportional to one quarter of the RCS. However, there is no report regarding to the RCS as far as the authors know. This paper presents the RCS characteristics of 11 typical road debris for 79GHz-band radar available in automotive radar which is also compared with 24GHz-band.

This study reports the findings of resistivity surveys which were conducted at the initiation area of debris flow at KM 9, Fraser’s Hill Gap road (FT148). The study involves three slope parallel survey lines and two lines perpendicular to the slope face. The parallel lines are FH01, FH02, and FH03, while the lines FH04 and FH05 are perpendicular. A granite body was detected at the central part of the east line and is nearest to the ground surface along FH02. The existence of low resistivity zones within the granite body is interpreted as highly fractured, water conducting zones. These zones are continuous as they have been detected in both the east-west as well as the north-south lines. The residual soil layer is relatively thin at zones where weathered granite dominates the slope face of the failure mass. The weak layer is relatively thick with an estimated thickness of 80 m and water flow occurs at the base of it. The high water flow recorded from the horizontal drains further supports the possible existence of these highly fractured, water conducting zones located within the granite. The shallow fractured granite is virtually “floating” above the water saturated zone and therefore is considered unstable.

Frequency-diversity is a computational imaging technique that can offer all-electronic imaging systems by leveraging spatio-temporally incoherent radiation patterns as an enabling technology. This approach exhibits a significant contrast to conventional imaging modalities, such as synthetic aperture radars (SAR) and phased arrays in that the raster scanning requirement of a scene to be imaged (mechanical or electronic) can be broken and replaced by a quasi-random interrogation of the scene. This aspect of frequency-diverse computational imaging systems significantly simplifies the physical hardware requirements of conventional radars. Despite this advantage, the application of the frequency-diversity technique has been mostly limited to static imaging scenarios, where the position of the scene to be imaged remains fixed over the data acquisition cycle. This limitation hinders the frequency-diverse computational radars from being deployed for applications where the scene dynamics may vary over the data acquisition cycle, such as in automotive radars. In this paper, we demonstrate that by modifying the sensing matrix to account for the movement of the radar platform, frequency-diverse computational imaging radars can be successfully used in debris detection on roads. We show that operating within the frequency band of 77-81 GHz, the presented dynamic frequency-diverse radar technique can produce high fidelity point spread function (PSF) patterns eliminating the distortions caused by the motion of the radar. We also prove that the PSF patterns of the radar are in excellent agreement with theoretical diffraction limited resolution limits.

This dissertation comprises two parts: Part I, which discusses the growing trend in project books in contemporary poetry, and Part II, a collection of poems titled, Road Debris. There is an increasing trend in the number of project books, which are collections of poetry unified in both thematic and formal ways. the individual poems in a project book share overt connections which allow the book to work on many different levels, blending elements of fiction and non-fiction or sharing a specific theme or speaker. While these books have the advantage of being easily memorable, which might gain poets an edge in book contests, there are also many risks involved. the main issue surrounding project books is if the individual poems can justify the book, or do they seem too repetitive or forced. As more poets, especially newer ones, try to use the project book as a shortcut to publication, it can result in poorly written poems forced to fit into a particular concept. By examining three successful cotemporary project books—The Quick of It, by Eamon Grennan; Incident Light, by H. L. Hix; and Romey’s Order by Astory Riley—this essay discusses how these books work in order to understand the potential of the project book. All of these books work in distinctly different ways, yet they all fall into the category of project book. While project books will inevitably result in poor imitations, it allows books of poetry to expand and explore in different directions.

Abstract The road along the Óshlíð hillside in the West Fjords region of Iceland is one of the most hazardous roads in Iceland due to avalanches, rockfalls and debris flows. The road has little traffic, but nevertheless traffic accidents caused by the severe conditions at the site are common. A number of avalanche tracks are found on the hillside. In some of these tracks, avalanches occur more frequently than in others. When there is an avalanche threat, avalanches generally flow over many tracks in a short time. Monitoring vibrations in the tracks with the highest avalanche frequency can indicate when avalanches start flowing down the hillside in a snowstorm, and avalanche hazard can then be declared with the specific site indicated. The same methodology can be used for rockfalls and debris flows, which are strongly affected by weather conditions and typically occur in clusters. Based on this knowledge, a research project was initiated in February 1996 with the objective of developing an automatic system based on seismic measurements to detect and analyze avalanches on the Óshlíð hillside and to instantly send a warning to a control station.