ACF object detector for FPGA

ACF object detector for FPGA

Software

IP core for object detection in image using ACF detector. This uses multi-channel image features and a cascaded classifier with a decision tree. During evaluation, it is possible to parallelize the computation by partitioning the features into banks, thus increasing the performance of the system linearly, but this must be taken into account during the training process. In the configuration file, it is possible to set IP core parameters such as the maximum resolution of the input image, the number of classifiers and their length, the number of banks to be evaluated in parallel, the size range of the detected objects, etc., to influence the performance and the amount of resources consumed. The Detection IP core can process video up to 4K resolution, classify up to 8 different object classes, and provides sufficient performance to process FullHD video at 60 fps. 

IP core for object detection in image using ACF detector. This uses multi-channel image features and a cascaded classifier with a decision tree. During evaluation, it is possible to parallelize the computation by partitioning the features into banks, thus increasing the performance of the system linearly, but this must be taken into account during the training process. In the configuration file, it is possible to set IP core parameters such as the maximum resolution of the input image, the number of classifiers and their length, the number of banks to be evaluated in parallel, the size range of the detected objects, etc., to influence the performance and the amount of resources consumed. The Detection IP core can process video up to 4K resolution, classify up to 8 different object classes, and provides sufficient performance to process FullHD video at 60 fps. 

Object detection, ACF, FPGA, decision tree

Object detection, ACF, FPGA, decision tree

Místnost Q201, Fakulta informačních technologií VUT v Brně, Božetěchova 2, 612 66 Brno

K využití výsledku jiným subjektem je vždy nutné nabytí licence

https://www.fit.vut.cz/research/product/687/