FPGA based Deterministic Latency Image Acquisition and Processing System for Automated Driving Systems
| Autor: | Alexander Warren, Jahanzeb Ahmad |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Computer science
business.industry Deep learning Real-time computing 02 engineering and technology Convolutional neural network 020202 computer hardware & architecture Kernel (image processing) 0202 electrical engineering electronic engineering information engineering Image acquisition 020201 artificial intelligence & image processing Artificial intelligence Central processing unit Latency (engineering) Image sensor Field-programmable gate array business |
| Zdroj: | ISCAS |
| DOI: | 10.1109/iscas.2018.8351472 |
| Popis: | Automated driving systems are one of the key motive in latest developments in artificial intelligence and hardware technologies required for artificial intelligence. These systems have a sheer number of sensors including image sensors. These image sensors produce a huge amount of visual data. This data is generally processed using traditional image recognition and classification techniques or the latest deep learning techniques like Convolutional Neural Networks (CNNs). The accuracy and latency of the systems have great impact on how the vehicle deals with its surroundings. In this paper, we discuss how Field-Programmable Gated Array's (FPGA) flexible input/output structure enables us to implement a deterministic latency acquisition and processing system. Experimental results show that by bypassing the host CPU from the input data path overall system becomes orders of magnitude more deterministic. When the CPU is bypassed from the input data path, non-determinism in overall processing is up-to 115.5 nsec as compared to 77.4 msec when CPU is in the input path. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|