Efficient Real-Time Object Detection based on Convolutional Neural Network

Visual objects recognition based on deep learning with the CNN (Convolutional Neural Network) have been extensive developments in computer image algorithms and improvements in computer hardware. In this work, a simplified CNN structure with a simple hardware system based on the Raspberry Pi with a compatible camera is developed. It can take pictures or videos in real time to recognize the existence of a human among ten predefined classes. The proposed model is implemented with less memory and less processing power while handling large amounts of data with Pascal VOC and Microsoft COCO datasets. The detection algorithm can quickly distinguish the object in 0.82 seconds and higher. The accuracy reaches 98%.