研究目的
To develop a low-cost camera-based algorithm for vehicle detection, tracking, and counting to manage traffic flow and reduce congestion and accidents.
研究成果
The proposed algorithm effectively detects, tracks, and counts vehicles with high accuracy, as shown by IoU comparisons with ground truth. It performs well in both daytime and nighttime conditions and is more efficient than some existing methods. Future work could involve implementation with OpenCV for online processing and improvements to handle limitations such as shadows and fast-moving vehicles.
研究不足
Vehicle shadows are included in the foreground, non-stationary cameras are not handled, high vehicle speeds can reduce accuracy, and intense sunlight causes reflections that add noise. These issues could be addressed with techniques like thresholding or Histogram of Oriented Gradients (HOG).
1：Experimental Design and Method Selection:
The algorithm uses Gaussian Mixture Model for background subtraction to isolate vehicles, morphological operations for noise removal, BLOB analysis for vehicle detection, Hungarian algorithm for label association, and Kalman filter for tracking.
2：Sample Selection and Data Sources:
Videos were acquired from CCTV and IR cameras, including daytime and nighttime footage. Ground truth was self-constructed using annotation tools for some videos, and benchmark videos from UCSD database were used for comparison.
3：List of Experimental Equipment and Materials:
CCTV camera, IR camera, MATLAB 2017a software.
4：Experimental Procedures and Operational Workflow:
Input video frames are processed through background subtraction, morphological operations, BLOB analysis, Hungarian algorithm for counting, and Kalman filter for tracking, with a region of interest (ROI) defined to focus on vehicles entering the detection field.
5：Data Analysis Methods:
Accuracy was evaluated using Intersection Over Union (IoU) method to compare predicted bounding boxes with ground truth, and results were compared with existing methods.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
