A Color Histogram Based Large Motion Trend Fusion Algorithm for Vehicle Tracking
Open Access
- 9 June 2021
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 83394-83401
- https://doi.org/10.1109/access.2021.3087904
Abstract
Due to the static nature of roadside cameras, targets of images have the characteristics of near big and far small. In a short time, as distances between targets and cameras get more faraway and the pixel ratios of the same target decrease sharply, which leads to a decrease in appearance information. However, the loss of appearance information makes tracking more challenging in vehicle tracking. To solve this problem, we propose a color histogram based large motion trend fusion (LMT-CH) algorithm for vehicle tracking, suitable for the traffic scene taken by the roadside camera. Based on Generalized Intersection-over-Union and Hungarian algorithm, LMT-CH uses targets of the large motion trend combined with the color histogram to modify the associated results. The algorithm includes four key steps: 1) train model based on the EfficientDet framework as an object detector. 2) First, calculate the correlation matrix between two consecutive frames. Second, get the matching results based on the Hungarian algorithm. Then, divide targets into four different sets based on the matching results. 3) Calculate the color histogram and large motion trend of the target in the suspected increase set of current frame's targets. 4) Re-identification trajectories based on large motion trend and color histogram. We evaluate the algorithm for the UA-DETRAC datasets with different levels of congestion. In several experiments, we show that the algorithm can effectively deal with the problem of target occlusion and loss, and it performs well in the ID switch of long-distance targets.Funding Information
- Beijing Nova Program of Science and Technology (Z191100001119087)
- Beijing Municipal Science and Technology Commission (Z181100004618005)
This publication has 26 references indexed in Scilit:
- High-Speed tracking-by-detection without using image informationPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2017
- Simple online and realtime trackingPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2016
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksIEEE Transactions on Pattern Analysis and Machine Intelligence, 2016
- Deep Relative Distance Learning: Tell the Difference between Similar VehiclesPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2016
- Deep Residual Learning for Image RecognitionPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2016
- Fast R-CNNPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2015
- Rich Feature Hierarchies for Accurate Object Detection and Semantic SegmentationPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2014
- Histograms of Oriented Gradients for Human DetectionPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2005
- A New Approach to Linear Filtering and Prediction ProblemsJournal of Basic Engineering, 1960
- The Hungarian method for the assignment problemNaval Research Logistics Quarterly, 1955