Target tracking method based on Deepsort algorithm

• Main Authors: GUO LUAN, YAO YAO, CHANG XIAOXU, ZHOU BICHEN, HAN CHAO, CHEN MENGDA, LYU QINGBIN, LIU JIAJING, SHA SHUO, GUO ZIJUN, WU YANG, DAI JINZHOU, ZHANG LIN
• Format: Patent
• Language: Chinese; English
• Published: 31.05.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention discloses a target tracking method based on a Deepsort algorithm, and relates to the technical field of target tracking, and the method comprises the steps: capturing a vehicle image sequence, carrying out the vehicle detection, and extracting a vehicle image region; extracting a feature vector from the vehicle image region; predicting and updating the motion state of the vehicle, and carrying out adaptive similarity measurement; inter-frame dynamic target association is carried out according to the predicted vehicle motion state in combination with a similarity measurement result; and carrying out vehicle re-identification and continuous tracking. The adaptive similarity measurement is introduced, the similarity calculation mode is dynamically adjusted according to the specific condition of the real-time scene, the accuracy and robustness of target tracking in the complex scene are improved, the challenges of shielding and view angle change are effectively dealt with, and by combining the desig