A Comparative Study of State-of-the-Art Deep Learning Algorithms for Vehicle Detection

This paper introduces a novel low-cost solar-powered wearable assistive technology (AT) device, whose aim is to provide continuous, real-time object recognition to ease the finding of the objects for visually impaired (VI) people in daily life. The system consists of three major components: a miniature low-cost camera, a system on module (SoM) computing unit, and an ultrasonic sensor. The first is worn on the user’s eyeglasses and acquires real-time video of the nearby space. The second is worn as a belt and runs deep learning-based methods and spatial algorithms which process the video coming from the camera performing objects’ detection and recognition. The third assists on positioning the objects found in the surrounding space. The developed device provides audible descriptive sentences as feedback to the user involving the objects recognized and their position referenced to the user gaze. After a proper power consumption analysis, a wearable solar harvesting system, integrated with the developed AT device, has been designed and tested to extend the energy autonomy in the different operating modes and scenarios. Experimental results obtained with the developed low-cost AT device have demonstrated an accurate and reliable real-time object identification with an 86% correct recognition rate and 215 ms average time interval (in case of high-speed SoM operating mode) for the image processing. The proposed system is capable of recognizing the 91 objects offered by the Microsoft Common Objects in Context (COCO) dataset plus several custom objects and human faces. In addition, a simple and scalable methodology for using image datasets and training of Convolutional Neural Networks (CNNs) is introduced to add objects to the system and increase its repertory. It is also demonstrated that comprehensive trainings involving 100 images per targeted object achieve 89% recognition rates, while fast trainings with only 12 images achieve acceptable recognition rates of 55%.

show abstract

“…Nevertheless, our mAP agrees with those reported in [70,71]. In the first, an 81.09% mAP was obtained for moving vehicle detection.…”

Section: Results Discussionsupporting

confidence: 92%

Solar-Powered Deep Learning-Based Recognition System of Daily Used Objects and Human Faces for Assistance of the Visually Impaired

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Our results show the value of DetectNet as an object detection model based on coverage maps rather than the anchor-based models such as Faster R-CNN, SSD, and R-FCN [76]. Although they share the same backbone architecture in some cases, the bounding box extraction strategy may show different performance for specific scenarios [77]. However, further studies are needed to understand the effects of deep learning architectures in the generation of covering maps [78].…”

Section: Discussionmentioning

confidence: 99%

Assessment of Tree Detection Methods in Multispectral Aerial Images

et al. 2020

View full text Add to dashboard Cite

Detecting individual trees and quantifying their biomass is crucial for carbon accounting procedures at the stand, landscape, and national levels. A significant challenge for many organizations is the amount of effort necessary to document carbon storage levels, especially in terms of human labor. To advance towards the goal of efficiently assessing the carbon content of forest, we evaluate methods to detect trees from high-resolution images taken from unoccupied aerial systems (UAS). In the process, we introduce the Digital Elevated Vegetation Model (DEVM), a representation that combines multispectral images, digital surface models, and digital terrain models. We show that the DEVM facilitates the development of refined synthetic data to detect individual trees using deep learning-based approaches. We carried out experiments in two tree fields located in different countries. Simultaneously, we perform comparisons among an array of classical and deep learning-based methods highlighting the precision and reliability of the DEVM.

show abstract

“…The vibration information obtained during the operating period of the IWM represents its operating conditions. Therefore, it is vital to select and extract highly sensitive SPs from the vibration signals for fault diagnosis and fault-type recognition [35]. However, the mechanical fault features of the IWM are often concealed by the effect of different speeds and different road levels; therefore, it is necessary to conduct a comprehensive analysis of the normal and abnormal states in order to obtain highly sensitive SPs.…”

Section: Selection Of Highly Sensitive Sps For Fault Diagnosismentioning

confidence: 99%

Real-Time Diagnosis of an In-Wheel Motor of an Electric Vehicle Based on Dynamic Bayesian Networks

et al. 2019

View full text Add to dashboard Cite

This study proposes a new real-time diagnosis method for an in-wheel motor (IWM) of an electric vehicle (EV) based on dynamic Bayesian networks (DBNs). Since the electrical signal of the vehicle power supply is unstable because of the interference resulting from the EV's frequent acceleration and deceleration, the IWM's vibration signal is focused. Symptom parameters (SPs) in the time and frequency domains are used to represent different features of the vibration signals in the actual operating conditions of the EV. To select highly sensitive SPs, stable average discrimination rate (SADR) is proposed, which consists of the average discrimination rate (ADR) and the stability coefficient of the group (SCG). Moreover, DBNs are employed to establish a model for the real-time diagnosis of the IWM's mechanical faults, in which the parameter of road-speed-time slice (RSTS) is used to solve the problem that the state transition probability distribution between two continuous time slices cannot be obtained. Finally, the effectiveness of the proposed methods is verified by experiments using the IWM test bench. INDEX TERMS Dynamic Bayesian networks, electric vehicle, in-wheel motor, real-time diagnosis, roadspeed-time slice.

show abstract

A Comparative Study of State-of-the-Art Deep Learning Algorithms for Vehicle Detection

Cited by 117 publications

References 30 publications

Solar-Powered Deep Learning-Based Recognition System of Daily Used Objects and Human Faces for Assistance of the Visually Impaired

Solar-Powered Deep Learning-Based Recognition System of Daily Used Objects and Human Faces for Assistance of the Visually Impaired

Assessment of Tree Detection Methods in Multispectral Aerial Images

Real-Time Diagnosis of an In-Wheel Motor of an Electric Vehicle Based on Dynamic Bayesian Networks

Contact Info

Product

Resources

About