Project Details

INSEM - Camea SV

Project Period: 1. 7. 2022 - 31. 5. 2023

Project Type: contract

Partner: CAMEA, spol. s r. o.

English title
Intelligent Sensors for Traffic Monitoring - Camea SV
Type
contract
Keywords

intelligent sensor, radar, edge computing, cloud computing

Abstract

1. Stage of exploring the possibilities of merger and preparation of the initial set of procedures

1. 7. 2022 30.11.2022 Exploring the possibility of processing the fusion of camera data (image) and sensor data obtained from a 3D radar sensor (point cloud) and processing procedures for such data fusion. The aim is in particular:

Prepare, in cooperation with CAMEA (and another cooperating company COGNITECHNA), a dataset of traffic scenes with image and radar data as well as with annotations (at least 500 shots). Prepare the structure of the neural network and prepare the input data so that it is compatible with the network for the purpose of data analysis for object detection, classification and categorization. Prepare algorithms implementing data analysis by neural network (CNN type), select a suitable library for this software to run in the smart camera system CAMEA (based on Intel Atom). Experimentally implement the initial version of the network configuration and verify compatibility and basic functionality (for example, on vehicle detection).

2. Stage of merger optimization and processing of a verified set of procedures

1. 12. 2022 31.5.2023 Optimization of fusion parameters and delivery of an updated version of the relevant set of procedures for image and 3D data fusion (point cloud). The aim is in particular:

Optimize the detection procedure itself so that it detects objects in traffic with a reliability of at least 90% (exact metrics and categories of objects will be specified during the 1st stage). Accelerate detection so that real-time detection takes place on the smart camera system (faster than the flow of objects, the exact parameters will be specified in the 1st stage). Prepare procedures implementing data analysis by neural network and prepare it for operation both on the collected dataset and in the conditions of real data entry. Verify the prepared optimized software in laboratory conditions close to real traffic (from the point of view of algorithms and data flow, not necessarily in deployment directly in road traffic).

Team members
Hajduk Petr, Ing. (Děkanát FIT VUT) , research leader
Zemčík Pavel, prof. Dr. Ing. (UPGM FIT VUT) , team leader
Kula Michal, Ing., Ph.D. (UPGM FIT VUT)
Nesvedová Šárka (Děkanát FIT VUT)
Back to top