In the AICUP 2024 “Multi-Camera Multi-Target Vehicle Tracking Challenge,” the ACVLab team received both the Merit Award and the Creative Execution Award, showcasing exceptional AI model design capabilities and innovative application potential. As a flagship AI technology competition platform in Taiwan, AICUP promotes industry-academia collaboration and practice-oriented technology development. This year's competition focused on the challenging and highly applicable topic of multi-camera vehicle tracking in smart transportation.
To tackle the technical difficulty of accurately tracking the same vehicle across multiple camera perspectives, ACVLab developed an unsupervised deep learning framework that does not rely on training data. This framework effectively performs vehicle re-identification (Re-ID) and trajectory linking, overcoming the limitations of traditional surveillance systems caused by restricted fields of view. Through innovative model architecture and flexible application strategies, ACVLab significantly improved the accuracy and efficiency of cross-camera tracking, demonstrating strong feasibility for real-world deployment.