Key Features
- Advanced HOG/SVM object classification core for support of Multi-Object detections in camera-based video systems
- Supports AMD Zynq™ 7000 SoC, Zynq™ UltraScale+™ and FPGAs
- Up to 4 SVM engines in a parallel
- ARM® AMBA® AXI4-Stream and AXI4-Lite compliant
- Objects models loadable run-time via software
- Maximum video input 2048x2048
- Support for multiple scale detection
- Run-time variable image size
- High Bandwidth > 600 Mpixel/sec
- High Throughput; >9.3 classification/sec, > 38.4 GMAC/sec
- Low latency (< 8 lines)
- Pedestrian classifier trained on wide range of automotive scenarios
- Prepackaged for the AMD Vivado™ Design Suite
- Supported by advanced software training kits
Description
The logiHOG is an HOG/SVM-based object detection logicBRICKS IP core for detection of multiple objects in vision-based embedded applications. It is a direct successor of the logiPDET Pedestrian Detector IP core. The algorithm follows a discriminative approach by combining the HOG-based descriptor and the SVM classifier. HOG (Histogram of Oriented Gradients) is a descriptor designed to encode object structure. It is particularly suitable for pedestrian detection. SVM (Support Vector Machine) is a non probabilistic binary linear classifier. The logiHOG IP core streams out the input frames from the external memory, internally generates a pyramid of images at different scales, and finally classifies a detection window sliding over the entire pyramid in order to detect differently sized objects, or the objects moving in an arbitrary range. Depending on the trained model, the logHOG may detect different objects, such as pedestrians, vehicles, traffic signs, faces, etc.
| The logiHOG core is sourced from Technology Partner
EVS Embedded Vision Systems Srl. |