基于事件相机与事件-信号-图像转换的市政道路病害检测方法OA

The municipal road diseases detection methodologies based on digital imagery and 3D laser scanning are prone to interference from complex environmental conditions such as intense illumination and shadowing,leading to compromised accuracy.To address this problem,a novel approach for municipal road distress detection and mor-phological segmentation utilizing event cameras coupled with event-signal-image conversion is introduced in this paper.Firstly,an event camera-a high-frequency dynamic sensing device-is employed to capture event data of the pavement surface,which are represented as event feature units,enabling road surface information collection in complex municipal road environments of strong light and shadows.Subsequently,an adaptive event feature accumu-lation algorithm compares the similarity of event feature units across consecutive timestamps and accumulates them into a distress feature matrix,ensuring effective aggregation of event-based road distress data.Finally,a short-time Fourier transform-based deep neural network is applied to automatically extract disease features from the feature matrix,accomplishing both diseases detection and morphological segmentation.Accuracy and stability tests were conducted under different sampling speeds and lighting conditions,involving a total of 1 745 municipal road in-spection datasets.Experimental results demonstrate that the method achieves recall,precision,F1-score,and Inter-section-over-Union(IoU)of disease symptom segmentation of 87.23％,87.28％,87.24,and 76.69％,respectively,significantly outperforming conventional digital image and 3D laser-based techniques.Moreover,the method main-tains consistent morphological segmentation accuracy under adverse conditions of intense light and shadows.