As a part of traffic surveillance use case, we integrate advanced HDR image capturing and object detection in FPGA, targeting Zynq platform in a smart camera. The camera will capture multi-exposure Full-HD video at 90 FPS and merge the image stream to 30 FPS tone-mapped HDR video using novel motion-aware merging algorithm.
In the traffic surveillance scenario we will detect license plates using our highly optimized object detector IP core. The detections along with the HDR image are made available for CPU part in order to execute other tasks like plate recognition or tracking. Such a technology will enable surveillance in challenging lighting conditions where other object detection methods often fails.
Currently our work focuses on improvements of the object detection core and its integration with the image merging algorithm and optimization of execution using event-based profiling.