Motion-based object detection method, object detection apparatus and electronic device

Abstract

A motion-based object detection method includes the steps of extracting, by processing acquired first and second images, one or more regions of interest (ROIs); transforming the one or more ROIs intograyscale; and acquiring, by processing the grayscale ROIs with a deep neural network (DNN) model to classify the objects contained in the one or more ROIs, a classification result of whether the objects contained in the one or more ROIs belong to a given categories. The DNN model comprises N (N is a positive integer and ranged from 4-12) depthwise separable convolution layers; each depthwise separable convolution layer comprises a depthwise convolution layer for applying a single filter to each input channel and a pointwise layer for creating a linear combination of the outputs of the depthwise convolution layer to obtain feature maps of the grayscale ROIs.

Description

technical field [0001] The present invention generally relates to machine vision, and in particular to a motion-based object detection method, a machine object detection device and electronic equipment, which can be applied to vision applications on an embedded platform for real-time object detection, and have lower performance consumption and an effective balance between latency and accuracy. Background technique [0002] Humans can quickly and accurately identify the category an object belongs to through domain knowledge. In the age of information technology, automatic object recognition using machine vision has become a common demand. For example, surveillance camera devices, which can be used to identify potential intruders by distinguishing between specific objects (eg, people) and inanimate objects (eg, scene backgrounds) through the integration of object recognition computer programs. [0003] In recent years, deep neural networks (e.g., convolutional neural network...

AI Technical Summary

Problems solved by technology

[0004] Such a development trend is not conducive to the promotion of deep learning networks in embedded terminals. The reasons are as follows: First, the computing performance of embedded chips in embedded terminal products (such as monitoring equipment) is not very strong. Computing still needs to consume a lot of bandwidth resources and computing resources; second, for embedded terminal products, the requirements are low latency and low power consumption, and at the same time, it only needs to meet the accuracy within an acceptable range

Images