Volume 9, Number 17, December 2019

Compressed Video Stream Based Object Detection

  Authors

Pyeong Kang Kim¹, Hyung Heon Kim¹, Tae Woo Kim¹ And Young Kyun Cha², ¹Innodep.Inc, Korea and ²Korea University, Korea

  Abstract

Nowadays, the need for research on an intelligent video monitoring system is increasing worldwide. Among the object detection methods, the core technology of the intelligent video monitoring system, or object detection using a deep learning-based convolutional neural network, is used widely due to its proven performance. Nonetheless, deep learning-based object detection requires many hardware resources because it decodes the videos to analyze. Therefore, this article suggests an advanced object recognition technique by conducting compressed video stream-based object detection in order to reduce consumption of resources for object detection as well as improve performance and confirms via the performance evaluation that speed and recognition rate improved compared to existing algorithms such as YOLO, SSD, and Faster RCNN.

  Keywords

object detection, convolutional neural network, Inception V4, Motion Vector

Conference Proceedings