Loading…

Pedestrian and Multi-Class Vehicle Classification in Radar Systems Using Rulex Software on the Raspberry Pi

Nowadays, cities can be perceived as increasingly dangerous places. Usually, CCTV is one of the main technologies used in a modern security system. However, poor light situations or bad weather conditions (rain, fog, etc.) limit the detection capabilities of image-based systems. Microwave radar dete...

Full description

Saved in:
Bibliographic Details
Published in:Applied sciences 2020-12, Vol.10 (24), p.9113
Main Authors: Daher, Ali Walid, Rizik, Ali, Randazzo, Andrea, Tavanti, Emanuele, Chible, Hussein, Muselli, Marco, Caviglia, Daniele D.
Format: Article
Language:English
Subjects:
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Nowadays, cities can be perceived as increasingly dangerous places. Usually, CCTV is one of the main technologies used in a modern security system. However, poor light situations or bad weather conditions (rain, fog, etc.) limit the detection capabilities of image-based systems. Microwave radar detection systems can be an answer to this limitation and take advantage of the results obtained by low-cost technologies for the automotive market. Transportation by car may be dangerous, and every year car accidents lead to the fatalities of many individuals. Humans require automated assistance when driving through detecting and correctly classifying approaching vehicles and, more importantly, pedestrians. In this paper, we present the application of machine learning to data collected by a 24 GHz short-range radar for urban classification. The training and testing take place on a Raspberry Pi as an edge computing node operating in a client/server arrangement. The software of choice is Rulex, a high-performance machine learning package controlled through a remote interface. Forecasts with a varying number of classes were performed with one, two, or three classes for vehicles and one for humans. Furthermore, we applied a single forecast for all four classes, as well as cascading forecasts in a tree-like structure while varying algorithms, cascading the block order, setting class weights, and varying the data splitting ratio for each forecast to improve prediction accuracy. In the experiments carried out for the validation of the presented approach, an accuracy of up to 100% for human classification and 96.67% for vehicles, in general, was obtained. Vehicle sub-classes were predicted with 90.63% accuracy for motorcycles and 77.34% accuracy for both cars and trucks.
ISSN:2076-3417
2076-3417
DOI:10.3390/app10249113