An Image Recognition System for Multiple Video Inputs over a Multi-FPGA System

A multi-user shared FPGA-based cloud computingplatform is presented for an image recognition application withmultiple video inputs. The platform is made of a sea of FPGAdevices, each connected as a hierarchical ring network. The users place IP cores called tiles on the platform and connect the tiles to each other to form a sequence. An architecture featuring aparallel pipeline of tiles and an intra-tile seIP cores called tiles on the platform and connect the tiles to each other to form a sequence. An architecture featuring aparallel pipeline of tiles and an intra-tile selector network isadopted so that users can work their own tile sequences in parallelwithout causing any interference with other users. A set ofapplication program interfaces (APIs) for writing tiles, deletingtiles, and configuring the intra-tile selector network isimplementedlector network isadopted so that users can work their own tile sequences in parallelwithout causing any interference with other users. A set ofapplication program interfaces (APIs) for writing tiles, deletingtiles, and configuring the intra-tile selector network isimplemented. With the APIs, the proposed system manipulatesthe tiles on FPGAs in response to users´ demands. The platform issuitable for running applications requiring low-latency andhigh-throughput data-processing capability. A histograms oforiented gradients feature extractor and Real AdaBoost classifierare implemented to perform image recognition.