A CNN-based path trajectory prediction approach with safety constraints

Mostafa Zaman, Nasibeh Zohrabi, Sherif Abdelwahed

Research output: Chapter in Book/Report/Conference proceedingConference contribution

8 Scopus citations

Abstract

Safety is one of the essential aspects to be considered in operating self-driving vehicles. In this paper, we aim to detect lanes and vehicles from input video or image by implementing advanced image thresholding techniques to detect lanes. We also use a linear support vector machine (SVM) classifier to detect vehicles from the image or video. A convolutional neural network (CNN) uses the images with the specified lane and vehicle detection as an input. It predicts the intention of going right, left, or drive straight based on the relative car distance on the video or the image. The main idea is to integrate CNN with lane and vehicle detection modules to estimate a safety path progression for a specific amount of time from the video or image based on the relative distance from other vehicles. Simulation results are given to illustrate the effectiveness of the proposed approach.

Original languageEnglish (US)
Title of host publication2020 IEEE Transportation Electrification Conference and Expo, ITEC 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages267-272
Number of pages6
ISBN (Electronic)9781728146294
DOIs
StatePublished - Jun 2020
Event2020 IEEE Transportation Electrification Conference and Expo, ITEC 2020 - Chicago, United States
Duration: Jun 23 2020Jun 26 2020

Publication series

Name2020 IEEE Transportation Electrification Conference and Expo, ITEC 2020

Conference

Conference2020 IEEE Transportation Electrification Conference and Expo, ITEC 2020
Country/TerritoryUnited States
CityChicago
Period6/23/206/26/20

All Science Journal Classification (ASJC) codes

  • Energy Engineering and Power Technology
  • Automotive Engineering
  • Transportation

Fingerprint

Dive into the research topics of 'A CNN-based path trajectory prediction approach with safety constraints'. Together they form a unique fingerprint.

Cite this