2014/07/23 by Philip Lenz, Andreas Geiger, Lenz, Philip +3
Computer Science · Decision Sciences · #Advanced Bandit Algorithms Research #Anomaly Detection Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #Data Stream Mining Techniques #FOS: Computer and information sciences #Security in Wireless Sensor Networks #Video Surveillance and Tracking Methods
paper · pdf · doi:10.48550/arxiv.1407.6251
openalex publication_date 2014/07/23 · openalex created_date 2022/10/05 · openalex updated_date 2026/07/28
One of the most popular approaches to multi-target tracking is\ntracking-by-detection. Current min-cost flow algorithms which solve the data\nassociation problem optimally have three main drawbacks: they are\ncomputationally expensive, they assume that the whole video is given as a\nbatch, and they scale badly in memory and computation with the length of the\nvideo sequence. In this paper, we address each of these issues, resulting in a\ncomputationally and memory-bounded solution. First, we introduce a dynamic\nversion of the successive shortest-path algorithm which solves the data\nassociation problem optimally while reusing computation, resulting in\nsignificantly faster inference than standard solvers. Second, we address the\noptimal solution to the data association problem when dealing with an incoming\nstream of data (i.e., online setting). Finally, we present our main\ncontribution which is an approximate online solution with bounded memory and\ncomputation which is capable of handling videos of arbitrarily length while\nperforming tracking in real time. We demonstrate the effectiveness of our\nalgorithms on the KITTI and PETS2009 benchmarks and show state-of-the-art\nperformance, while being significantly faster than existing solvers.\n