Addressing the issue of frequent identity switches in highway vehicle tracking caused by nonlinear vehicle motion and occlusion by other objects, this paper proposes a highway vehicle tracking method based on YOLOv7. This method first introduces the Efficient-Receptive Field Block Cross Stage Partial Concat module to enhance detection accuracy for vehicles of varying scales. Next, DeepSORT is integrated as a baseline model, utilizing the Extended Kalman Filter algorithm to predict the positions of vehicles exhibiting nonlinear motion. To further improve matching accuracy, the Distance Intersection over Union matching algorithm is introduced after cascade matching. Experimental results on the Inner Mongolia Autonomous Region highway vehicle tracking dataset demonstrate that the proposed vehicle tracking model reduces the number of identity ID switches by 46. This indicates that the proposed method effectively minimizes ID switches during vehicle tracking, achieving precise and stable highway vehicle tracking.
An Enhanced Highway Vehicle Tracking Method Based on Improved YOLOv7 and DeepSORT
08.11.2024
2058485 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Real-Time Detection and Tracking of Express Parcels Based on Improved YOLOv5+DeepSORT
Springer Verlag | 2023
|Freeway vehicle detection on improved YOLOv7
SPIE | 2023
|