Design Of Indoor Environmental Object Detection System Based On Cloud Platform

With the development of computer vision and deep learning,the assistive system based on computer vision for visually impaired people begins to sprout.At present,the object detection algorithm based on deep learning has made significant achievements,but it relies on the mighty computing power of GPU and cannot be applied in the assistive system for visually impaired people with limited hardware performance independently.Considering the contradiction between computing power and detection performance,this paper proposes an indoor environment object detection system based on the cloud platform.It has a strong theoretical and practical significance to complete a high-performance target detection system through the cooperation between intelligent terminals and cloud platform.The indoor environment object detection system based on cloud platform designed in this paper includes an intelligent terminal and a cloud platform.The intelligent terminal is composed of a camera,a display screen,an inertial measurement unit,a master control chip and other parts.It mainly realizes image collection,key-frame detection,saliency detection,image transmission and detection result display.The cloud platform is equipped with four Titan XP high-performance GPU graphics cards,mainly completing the object detection task based on deep learning.The higher resolution of the input image is,the better the object detection algorithm performs.Considering the contradiction between high resolution and low transmission efficiency,this paper proposes an improved region-based saliency detection algorithm for original image cutting.The saliency detection algorithm can be divided into two categories,fixed-point prediction and salient region prediction.The saliency detection algorithm based on salient region prediction is more suitable for the application scenario of image clipping.In this paper,we minimize the computational complexity of the saliency map by reducing the colors of the image.At the same time,the watershed segmentation algorithm is used for post-processing to get more precise salient regions for image clipping,to minimize the network transmission burden and improve the object detection performance.The object detection algorithm based on deep learning is divided into two types,one-stage and two-stage.The one-stage algorithm locates and recognizes objects directly.The two-stage algorithm extracts region proposals first and then classifies these region proposals.The one-stage object detection algorithm has high execution efficiency,but the detection accuracy needs to be improved.In this paper,an improved object detection algorithm is proposed.Based on YOLOv2,a more accurate network structure is designed to improve the accuracy of target detection.By using the deconvolution network to fuse the feature map of multiple layers,the representation ability of the feature maps of shallow layers is effectively improved.At the same time,on the basis of cross-entropy loss function,a more reasonable focus loss function is established to reduce the impact of imbalance of classification samples on model training.Through these improvements,a better object detection model is formed.Experimental results show that the proposed algorithm is superior to YOLOv2 in detecting accuracy,and the boundary boxes location is more accurate,but the execution efficiency is slightly reduced.In this paper,the saliency detection algorithm is transplanted to the intelligent terminal,and the object detection algorithm is transplanted to the cloud platform.The indoor environment object detection system based on cloud platform design is finally completed.Finally,the paper summarizes the research work carried out and looks forward to the future research direction.