基於深度學習之前方駕駛輔助移動物件行為分析系統

Front Moving Object Behavior Prediction System Exploiting Deep Learning Technology for ADAS Applications

蔡文嘉(Wen-Chia Tsai)

指導教授 ： 郭峻因

國立交通大學/電機學院/電子工程系所/碩士(2019年)

摘要

本論文基於C3D卷積行為辨識網絡來建構前方行人穿越預測系統。本論文嘗試改良了原始之C3D 3D卷積網絡，使得C3D行為辨識系統的網絡也能辨別目標物件的行為位置，此點有助於多重物件之行為分析辨識。並且，我們重新建構了原始C3D 3D卷積網絡架構的最後一層網路，使其學習於最新一幀中穿越之行人的位置資訊。
本系統除了於伺服器上開發之外，本論文亦將所提系統移植於低功耗嵌入式開發平台環境上執行並進行驗證。本論文執行於NVIDIA Jetson TX2開發平台上之行人穿越行為預測網絡，可達到7.5FPS之速度。

關鍵字

行人偵測 ； 深度學習 ； 行為辨識

並列摘要

This thesis proposed a front pedestrian crossing prediction system based on C3D convolution behavior prediction network. We improved the original C3D 3D convolution network to make behavior recognition network with low resolution input have the ability of object localization, which is important to detect multiple moving object behavior. Also we rebuilt the last layer of C3D 3D convolution layer to learn the crossing pedestrian location in the latest frame.
The proposed system is not only developed on servers but also implemented on the embedded systems. The proposed behavior prediction network can reach 7.5 FPS when it is implemented on NVIDIA Jetson TX2.

並列關鍵字

Pedestrian detection ； Deep Learning ； Behavior Recognition

參考文獻

[1] SIMONYAN, Karen; ZISSERMAN, Andrew. Two-stream convolutional networks for action recognition in videos. In: Advances in neural information processing systems. 2014. p. 568-576.

[2] DONAHUE, Jeffrey, et al. Long-term recurrent convolutional networks for visual recognition and description. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2015. p. 2625-2634.

[3] WANG, Limin, et al. Temporal segment networks: Towards good practices for deep action recognition. In: European conference on computer vision. Springer, Cham, 2016. p. 20-36.

[4] YAO, Li, et al. Describing videos by exploiting temporal structure. In: Proceedings of the IEEE international conference on computer vision. 2015. p. 4507-4515.

[5] FEICHTENHOFER, Christoph; PINZ, Axel; ZISSERMAN, Andrew. Convolutional two-stream network fusion for video action recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2016. p. 1933-1941.