An Multi-feature Fusion Object Detection System for Mobile IoT Devices and Edge Computing

With the increase of data scale and computing power, deep learning algorithm has made a prominent breakthrough in computer vision and other complex problems. However, its high complexity and large memory requirements make it very difficult to run in real time on the Internet of things terminal mobile devices. There is still delay the employing of cloud services cannot meet the real-time requirement. With the popularity of mobile terminal devices and the development of Internet of things, it is of great significance to design a real-time deep learning algorithm on IOT edge mobile devices with limited computing and memory resources. This paper proposes a new object detection method based on the current state-of-the-art object detection deep network model RetinaNet and traditional feature extraction method SIFT. RetinaNet is a one-stage detector with excellent detection speed and accuracy. We use RetinaNet as the object location method, then extract the CNN features and SIFT features of the fixed position image and combine them to train a new classifier. The object classification result will be based on the final classifier.