Design of Power Intelligent Safety Supervision System Based on Deep Learning
Bin Chen, Chen Hui, Kangli Zeng · 2018
In this paper, a real-time object recognition system based on deep learning is proposed to identify whether the operator has worn the specific safety equipment. Normally this is done by labor force to examine the real-time video feedback from camera, however it is extremely inefficiency and costly. Traditional object recognition methods are heavily relying on human designed features. As a result, their performance will downgrade dramatically under complex environment and cannot be used in real-time application due to low processing speed. In this paper, the original VGG16 neural network of Single Shot Multi-Box Detector (SSD) has been modified by introducing Inception module to improve its sensitivity to small objects, Combined with multi-object tracking technology to bind face recognition results with pedestrian identity, the experiments have proved that proposed system can recognize specific safety equipment of specific operator accurately and in real time at operation site.