APRIS2020 Special Technical Session
Starting at Nov. 9 ( 45min ) 12:45 (Thai), 13:45 (Malaysia), 14:45 (Japan)
Theme: Investigation of AI Techniques Based Object Detection on OpenCV on Inexpensive Embedded SoC
Speaker
QIANYU-ZHANG
Master of Electronics Engineering, 2020, PSU/Songklar.
Image processing on FPGA
TOYOTA TSUSHO NEXTY ELECTRONICS (THAILAND) CO,.LTD.
DEVICE SALES DIVISION, R&D SECTION, System Architect ( AI )
Abstract
Nowadays, the AI technique is more and more popular in our daily life, which has been widely used in many area, such as the object detection in the autonomous driving field. In our project, the object detection is deployed on the inexpensive embedded system on the forklift for detecting the people in the factory for decreasing the accidents. Whereas because of the requirement of both AI training and inference are high, the detecting time and the precision are still the key issues. Considering the real working environment, the fast speed and the high recognition can guarantee the workers’ safety. In the object detection field, the Yolo series models show the better performance than the others. In this project, different Yolo model such as Yolov3, Yolov3-tiny and Yolo-Lite are compared based on the detection precision and inference speed, in which the Yolov3-tiny is more potential because it’s faster than the Yolov3 and more correct than the Yolo-Lite. The pruned Yolov3-tiny that based on the yolov3-tiny showed the best performance that nearly same precision but faster speed than the Yolov3.
Facilitator
Masami IKURA
CTO
TOYOTA TSUSHO NEXTY ELECTRONICS (THAILAND) CO,.LTD.