Lightweight deep neural network method for personnel detection and people counting in elevator

Abstract

The invention belongs to the technical field of computer vision and target detection, and relates to a lightweight deep neural network method for personnel detection and people counting in an elevatorcar. According to the method, a convolutional neural network model is lightened, a Raspberry Pi 4B is taken as a development system, and embedded resources are fully utilized to locally realize personnel detection and people counting of an elevator car. According to the invention, a convolutional neural network (CNN) model structure comprises 11 blocks and totally comprises 23 convolution layers,region recommendation is carried out on an advanced semantic feature map, and the recommended region carries out dichotomy on a human head part and a background through a full connection layer; and on the basis of a multi-scale diversity target detection algorithm SSD, depth separable convolution is added to the convolution layer of each block of a network structure, the detection speed is obviously increased, video detection in a Raspberry Pi 4B system reaches FPS of 2.1, and the requirement that information feedback is smaller than 3S is met.

Description

technical field [0001] The invention belongs to the technical field of computer vision and target detection, and relates to a method for detecting people in an elevator car by using deep learning and embedded technology, in particular to a lightweight deep neural network for detecting people in an elevator car and counting people. network method. Background technique [0002] With the development of current social science and technology, many high-rise buildings and intelligent buildings have emerged. Correspondingly, a large number of vertical transportation tools - elevators have also emerged. Elevators are essential and important equipment to ensure the effective operation of high-rise buildings. At the same time, the safety of elevators is crucial to everyone who lives and works in high-rise buildings. Doing a good job in emergency rescue services is an elevator IoT company urgent need. At present, the technology based on personnel detection and people counting in the ...

AI Technical Summary

Problems solved by technology

[0007] To sum up, the main problems of the existing elevator car personnel detection and people counting technology are: 1. Infrared sensor perception technology: the detection accuracy is low when there are many passengers; 2. The elevator car based on traditional image processing People counting technology: background modeling has strict restrictions on the elevator car, and traditional image feature extraction has a low detection rate in complex environments; 3. The existing detection technology based on deep neural network, due to the huge amount of calculation Parameter training and inference platforms are mainly based on PCs and servers, and the investment cost of building and operating the platform is too high, especially for elevator car personnel detection and people counting, the high cost limits its application; 4. Existing Raspberry Pai's lightweight deep network image target detection method is less effective for identifying long-distance targets, such as head and shoulder targets with small corners

Images