Nowadays, urban road images contain a lot of information that can be detected and extracted. In such images, pedestrians, various types of motor vehicles and non-motor vehicles are the most important components as well as the focus of object detection. Object detection in road images is of great help to improve urban road management ability. By detecting road images, especially at intersections where traffic is often blocked, real-time statistics can be made on the traffic flow and crowd flow in that section. By uploading live road information to some navigation software or reporting it to a radio station, vehicles passing through the area can be notified to choose other roads. On the one hand, it saves the time of notified vehicles and minimizes the impact on the daily life of citizens. On the other hand, it avoids more vehicles going to the congested roads and aggravating the congestion, thus realizing the traffic dispatching without the influence of human factors such as traffic police. In addition, the traffic lights can also be coordinated with the real-time detection of vehicle and pedestrian flow to adjust the length of traffic lights to ensure the smooth road.
Object detection is the task of detecting different objects in images and videos. In this paper, a comprehensive review for the classical models is given first. Then the object detection performance in UAV images, as well as the design of lightweight and small-object detection models, are discussed as new directions for object detection.
R-CNN  applies convolutional neural networks to propose candidate regions for object localization and segmentation. This is followed by fine-tuning of specific areas to produce high performance improvements. The authors named the algorithm R-CNN because it combines the candidate regions with the convolutional neural network.
However, R-CNN also has some problems. Because the VGG16 network takes up a lot of space, training is expensive in space and time. Every candidate area performs Convolutional Network forward propagation, so the object detection process is slow.
- Operating system : Windows.
- Coding Language : Python.
System : Pentium IV 2.4 GHz or intel
Hard Disk : 40 GB.
Floppy Drive : 1.44 Mb.
Mouse : Optical Mouse.
Ram : 512 Mb.
Both the classical models and recent progress in object detection are discussed in this paper. We first discuss the main ideas of R-CNN, SSD, YOLO, etc. We also discuss the ideas of improving the object detection performance in UAV images, as well as the design of lightweight and small-object detection models. Our discussion is not only a look back of existing work, but also inspiring for future work.