Smart in-car camera system using mobile cloud computing framework for deep learning

Deep learning is becoming a popular technology in various applications, such as image recognition, gaming, information retrieval, for intelligent data processing. However, huge amount of data and complex computations prevent deep learning from being practical on mobile devices. In this paper, we designed a smart in-car camera system that utilizes mobile cloud computing framework for deep learning. The smart in-car camera can detect objects in recorded videos during driving, and can decide which part of videos needs to be stored in cloud platforms to save local storage space. The system puts the training process and model repository in cloud platforms, and the recognition process and data gathering in mobile devices. The mobile side is implemented in NVIDIA Jetson TK1, and the communication is carried out via Git protocol to ensure the success of data transmission in unstable network environments. Experimental results show that detection rate can achieve up to four frame-per-second with Faster R-CNN, and the system can work well even when the network connection is unstable. We also compared the performance of system with and without GPU, and found that GPU still plays a critical role in the recognition side for deep learning.