Portable system for automatic object tracking in video sequences
The portable system of automatic visual object tracking in a sequence of images of increased reliability and speed has been developed. The system finds the trajectory of a given object in video sequences determining its coordinates on each frame. The improvement of the system performance is achieved by developing and modifying tracking techniques as well as the structural and functional organization of the embedded software and hardware aids of their implementation on the ARM Cortex-A processor core. The composite method TLD and the method based on the discriminative correlation filter DCF have been improved that provided higher tracking speed on achromatic image sequences and allowed their software implementations to operate in a real-time in embedded computing devices. The new tracking method called FCT (Flow Clustering Tracker) has been developed. It is based on the clustering of the results of sparse optical flow estimation and affords the compromise between tracking robustness and accuracy when implementing it in portable solutions. The software implementations of the portable automatic tracking systems based on the PINE64 and Beagleboard-xM evaluation modules, which permit the real-time frame processing (at least 15 frames per second), are developed.