Embedded Computer Vision for Safety and Security:
an FPGA-based Approach for Detection and Classification
Automatic classifying different categories of objects in images and videos is one of the main goals in Computer Vision. Among them, classifying pedestrians has attracted considerable attention as a key component in different application domains such as video surveillance, navigation systems and robotic control.
Despite continuous efforts over recent years to improve accuracy and processing performance, they are not ready for real-world applications yet. Additionally, although hardware solutions have recently demonstrated their reliability to solve some problems in Computer Vision, few object detection systems are thought to be realized on embedded devices.
The recent development of new System on Chips (SoC), where processors, programmable logic and key peripherals are in a single device, brings new challenges by the possibility of non-serialized hardware/software partitioning. Moreover, it paves the way towards smarter and ubiquitous embedded vision systems capable of automatically detecting objects of interest directly in the field.
In this seminar, after a brief introduction of the PAVIS department, I will present our ongoing research towards developing robust pedestrian detection systems targeting surveillance and automotive applications. In particular, I will describe our FPGA-based pedestrian detection architecture using covariance matrices as object descriptor, and I will discuss recent advances in object detection using multi-cue features extracted from intensity, depth and motion.