The existing pedestrian target recognition algorithms often suffer from insufficient classification decision, which affects the accuracy of tracking. This paper proposes an algorithm that enhances tracking accuracy by incorporating target information from preceding and subsequent frames in the video. To tackle the issue of drift during target occlusion, the algorithm establishes a feature dictionary by extracting relevant information such as pedestrian contours, key points, and geometric features. Moreover, to address the issue of target drift, this algorithm employs a twin Network and a model update mechanism to process the filtered target information. Experimental results with measured data demonstrate that this algorithm outperforms traditional detection methods in characterizing local features of the target. Moreover, it successfully solves the challenges of classification decision and target drift in pedestrian recognition, offering improved stability and accuracy. This algorithm shows promising prospects for application in smart cities and other relevant domains. Future work will focus on conducting more optimizations to enhance adaptability to various environments and scenarios.
An Intelligent Pedestrian Tracking Algorithm Based on Sparse Models in Urban Road Scene
IEEE Transactions on Intelligent Transportation Systems ; 25 , 3 ; 3064-3073
2024-03-01
1287011 byte
Article (Journal)
Electronic Resource
English