Abstract:Unmanned aerial vehicles (UAV) play a pivotal role in the field of security owing to their flexibility, efficiency, and low cost. The realization of vehicle target detection, tracking, and positioning from the perspective of a UAV can effectively improve the efficiency of urban intelligent traffic monitoring. In this work, by fusing the target detection network, YOLO v4, with the detection-based multitarget tracking algorithm, DeepSORT, a method based on deep learning for automatic vehicle detection and tracking in urban environments, has been designed. With the aim of addressing the problem of UAV positioning a vehicle target, the state equation and measurement equation of the system have been constructed, and a particle filter based on interactive multimodel has been employed for realizing the state estimation of the maneuvering target in the nonlinear system. Results of the simulation show that the algorithm proposed in this work can detect and track vehicles automatically in urban environments. In addition, the particle filter algorithm based on an interactive multimodel significantly improves the performance of the UAV in terms of positioning the maneuvering targets, and this has good engineering application value.
computer science, information systems,telecommunications,engineering, electrical & electronic
What problem does this paper attempt to address?
This paper aims to solve the problems of detecting, tracking and locating vehicle targets by unmanned aerial vehicles (UAVs) in urban environments. Specifically, the paper proposes a deep - learning - based vision algorithm, which combines the YOLO v4 object detection network and the detection - based multi - object tracking algorithm DeepSORT, for automatically detecting and tracking vehicles in urban environments. In addition, in order to improve the performance of UAVs in locating maneuvering targets, the paper also constructs the state equation and measurement equation of the system, and adopts the interacting multiple model - based particle filter (IMM - PF) to achieve state estimation of maneuvering targets in nonlinear systems.
### Main Problems and Solutions
1. **Object Detection and Tracking**:
- **Problem**: Traditional traffic monitoring technologies such as induction coils, geomagnetic sensors and roadside cameras have problems such as small detection range, low accuracy and poor mobility, which limit the development of intelligent transportation systems.
- **Solution**: The paper proposes a deep - learning method combining YOLO v4 and DeepSORT for automatically detecting and tracking vehicles in urban environments. YOLO v4 has strong robustness and generalization ability, while DeepSORT improves the accuracy of multi - object tracking through Kalman filter prediction and Hungarian algorithm matching.
2. **Object Location**:
- **Problem**: For maneuvering targets, traditional static target location methods cannot meet the high - precision requirements because both the platform and the target are constantly moving and are subject to interference from multiple factors.
- **Solution**: The paper adopts the interacting multiple model - based particle filter (IMM - PF), which improves the tracking performance of highly maneuvering targets by fusing multiple motion models. The IMM - PF algorithm can handle nonlinear, non - Gaussian systems and provide better filtering effects.
### System Structure
- **Data Collection**: UAVs use cameras to shoot videos of the flight area and transmit the videos to the ground station through the data link.
- **Object Detection**: Vehicle target detection is carried out on the downloaded aerial videos at the ground station.
- **Object Tracking**: After vehicle targets are detected, mobile target tracking is carried out in subsequent video frames.
- **Object Location**: Extract the pixel coordinates of the target, combine the measurement data of the UAV's position, attitude and camera pointing angle, estimate the geographical coordinates of the target, and realize automatic detection, tracking and location based on vision technology.
### Key Algorithms
1. **YOLO v4 Object Detection Network**:
- Innovations include data augmentation, self - adversarial training, CSPDarknet53 backbone network, mish activation function, DropBlock, etc.
- Through the SPP module and FPN + PAN structure, the spatial resolution and context information of the feature map are enhanced.
2. **DeepSORT Vehicle Tracking Model**:
- Use Kalman filter prediction and Hungarian algorithm matching, combine Mahalanobis distance and cosine distance for information association and cascade matching, and improve the accuracy of multi - object tracking.
3. **IMM - PF Algorithm**:
- Combine the advantages of the interacting multiple model (IMM) and the particle filter (PF), and achieve high - precision location of highly maneuvering targets through recursive Bayesian filtering.
- It includes four steps: input interaction, model - matching filtering, model probability update and estimation output.
### Simulation Testing and Analysis
- **Dataset**: Use the urban traffic aerial images in the VisUAV multi - object aerial photography dataset, including 15,741 labeled vehicle images as training data.
- **Test Results**: Tests are carried out on the VisUAV2019 - MOT benchmark dataset. The results show that the algorithm can accurately detect and track multiple targets under different shooting angles, weather and lighting conditions, and significantly improve the UAV's performance in locating maneuvering targets.
In conclusion, this paper effectively solves the problems of detecting, tracking and locating vehicle targets by UAVs in urban environments by combining advanced deep - learning technologies and filtering algorithms, and has important engineering application value.