Deep learning has enabled a straightforward, convenient method of road pavement infrastructure management that facilitates a secure, cost-effective, and efficient transportation network. Manual road pavement inspection is time-consuming and dangerous, making timely road repair difficult. This research showcases You Only Look Once version 5 (YOLOv5), the most commonly employed object detection model trained on the latest benchmark Road Damage Dataset, Road Damage Detection 2022 (RDD 2022). The RDD 2022 dataset includes four common types of road pavement damage, namely vertical cracks, horizontal cracks, alligator cracks, and potholes. This paper presents an improved deep neural network model based on YOLOv5 for real-time road pavement damage detection in photographic representations of outdoor road surfaces, making it an indispensable tool for efficient, real-time, and cost-effective road infrastructure management. The YOLOv5 model has been modified to incorporate several techniques that improve its accuracy and generalization performance. These techniques include the Efficient Channel Attention module (ECA-Net), label smoothing, the K-means⁺⁺ algorithm, Focal Loss, and an additional prediction layer. In addition, a 1.9% improvement in mean average precision (mAP) and a 1.29% increase in F1-Score were attained by the model in comparison to YOLOv5s, with an increment of 1.1 million parameters. Moreover, a 0.11% improvement in mAP and 0.05% improvement in F1 score was achieved by the proposed model compared to YOLOv8s while having 3 million fewer parameters and 12 gigabytes fewer Giga Floating Point Operation per Second (GFlops).

The pedestrian attribute recognition task is becoming more popular daily because of its significant role in surveillance scenarios. As the technological advances are significantly more than before, deep learning came to the surface of computer vision. Previous works applied deep learning in different ways to recognize pedestrian attributes. The results are satisfactory, but still, there is some scope for improvement. The transfer learning technique is becoming more popular for its extraordinary performance in reducing computation cost and scarcity of data in any task. This paper proposes a framework that can work in surveillance scenarios to recognize pedestrian attributes. The mask R-CNN object detector extracts the pedestrians. Additionally, we applied transfer learning techniques on different CNN architectures, i.e., Inception ResNet v2, Xception, ResNet 101 v2, ResNet 152 v2. The main contribution of this paper is fine-tuning the ResNet 152 v2 architecture, which is performed by freezing layers, last 4, 8, 12, 14, 20, none, and all. Moreover, data balancing techniques are applied, i.e., oversampling, to resolve the class imbalance problem of the dataset and analysis of the usefulness of this technique is discussed in this paper. Our proposed framework outperforms state-of-the-art methods, and it provides 93.41% mA and 89.24% mA on the RAP v2 and PARSE100K datasets, respectively.

Road damage detection and classification is an important sector in computer vision and image processing. Road damage detection is rapidly getting a high application in various sectors of computer vision, especially in self-driving cars; it is getting great attention. We used the RDD-2020 dataset with four classes for training our models. Most of the existing methods using the RDD-2020 dataset failed to detect multiple overlapping damage regions of different classes in the same image. To overcome these limitations and enhance the previous model's performance, we experimented with five different pre-trained models. We fine-tuned the best-performing model (Faster RCNN with ResNet-101) by using three different optimizers (Adam, RMS Prop., and SGD with momentum) and kept batch sizes as 4, 8, 16, and 32. Then, we proposed a fine-tuned Faster RCNN with the ResNet-50 model. The experimented pre-trained models are EfficientDet, SSD with MobileNet-v1, SSD with MobileNetv2, SSD with ResNet-50, and Faster RCNN with ResNet-101. These models have a backbone network and a detecting head. The backbone network extracts the feature maps, and the detection head generates bounding box coordinates for detecting the damaged region and classifies these bounding boxes. We have found a maximum f1 score of 0.47 for the longitudinal class, 0.41 for transverse class, 0.41 for alligator class, and 0.35 for pothole using Fine-tuned Faster RCNN with ResNet-101 by keeping the batch size as 4 and optimizer as SGD with momentum which are better than the previous model of literature using RDD-2020 dataset.