High-speed Multiple Object Tracking Based on Fusion of Real-time Image Processing and Intelligent Image Processing
Tracking techniques are indispensable across various fields.
However, existing tracking systems often fail to meet all the criteria of low latency, high quality, and robustness across different working environments and target object types and quantities.
In response to this challenge, we have developed a tracking method designed to fulfill these criteria for objects with explicit templates and human joints without explicit templates.
Our approach integrates a hybrid tracking method with a label-based tracker management system for object tracking and dynamic search area setting method for skeleton tracking.
Our method has surpassed the conventional method, using Kalman filter prediction to match trackers, in RMSE and IOU metrics.
Also, we have demonstrated the capability to track 2 objects at 472 Hz and 4 objects at 178 Hz in object tracking and 6 joints at 380 Hz in skeleton tracking.
These results underscore the effectiveness and efficiency of our approach across various tracking scenarios.
Movie
If you want to use the original video, please send an e-mail for copyright permission to yamakawalabo
.