Skip to content

1 Introduction to Blindspot Assistance System

Jump to bottom
flaviojpriotti edited this page Jul 15, 2020 · 1 revision

OpenVINO Blindspot Assistance is a Computer Vision system based on Machine Learning that allows real time monitoring of objects behavior within the truck’s blindspots area, providing information about potential dangerous traffic situations.

This document defines the scope of the Object Detection Use case for the OpenVINO Blindspot Assistance System.

Narrative

As a driver, I want to have warnings on the cabin when people, objects or cars are too close to the vehicle and an attention mode when the vehicle is parking or backing up.

1st Stage Objectives

  1. To define the camera configuration for blind-spots monitoring a. Camera position calibration b. Detection area configuration c. Video input real-time processing.
  2. To validate the initial configuration for Object Detection.
  3. To validate target devices’ performances (CPU/GPU/MDX/NCS).
  4. To validate local alert Notification
  5. To validate system functionality in specific driving scenarios.

Assumptions

  1. Based on previous performance testing experiences, camera feeds will be processed individually and in parallel.
  2. There won’t be video signal synchronization
  3. The detection process will be tested only under daylight conditions
  4. Each event detected will be displayed in a simple notification window.
  5. The output for each one of the events detected will be transmitted to the cloud through MQTT communication protocol. No dashboard in the cloud.
  6. The notification system is considered an external service that will consume the detection output for displaying the proper warnings/notifications/alarms in a separate dedicated UI.
  7. All the detection models considered for the following POC are part of the OpenVINO toolkit. In case an open source model is found to provide wider range of objects detected and/or different lighting/weather conditions support, it will be evaluated in order to reduce the effort of training OpenVINO models.
  8. The range of objects able to be detected will be the following:
  • Cars
  • Trucks
  • Pedestrians
  • d. Bikes
  1. The initial scope is limited to Highway Driving scenario. These variations will be considered:
  • Forward driving
  • Lane Change
  • Left turn
  • Right turn
  1. CAN BUS integration is outside of the scope of this PoC.
Clone this wiki locally