A 201.4GOPS 496mW Real-Time Multi-Object Recognition Processor with Bio-Inspired Neural Perception Engine

The visual attention mechanism, which is the way humans perform object recognition [1], was applied to the implementation of a high performance object recognition chip [2]. Even though the previous chip achieved 50% gain of computational cost [2], it could recognize only one object in a frame so that it is not suitable for advanced multi-object recognition applications such as video surveillance, intelligent robots, and autonomous vehicle navigation [3]. A real-time multi-object recognition processor is presented based on the bioinspired visual perception algorithm. The proposed recognition processor has 4 features: 1) 3-stage pipelining with grid-based region-of-interest (ROI) processing for high recognition rate, 2) Neural perception engine (NPE) with three bio-inspired neural and fuzzy processing units for multi-object perception and segmentation, 3) Low latency multi-castable Network-on-Chip (NoC) for high bandwidth integration platform, and 4) Workload-aware power management for low power consumption.