Applied AI R&D

Patents & Publications

Patents (multi-national, worldwide applications)

  • US20230086009A1, WO2023049339A1 System and techniques to normalize objects in spatial imaging of spaces
  • US10970561B2 Detecting roadway objects in real-time images
  • US10586118B2 Localizing traffic situation using multi-vehicle collaboration
  • JP6729673B2 Location of traffic event by cooperation of multiple vehicles
  • US10395530B2 Situational understanding of unknown roadway conditions that are ahead for a connected vehicle
  • US10043084B2 Hierarchical context-aware extremity detection
  • JP6394735B2 Detection of limbs using hierarchical context-aware
  • US9805276B2 Generating real-time driver familiarity index for fine-grained dynamic road scenes
  • JP6341311B2 Real-time creation of familiarity index for driver’s dynamic road scene
  • US9792821B1 Understanding road scene situation and semantic representation of road scene situation for reliable sharing
  • JP6414255B Semantic representation of road scene situations for understanding and sharing of road scene situations
  • US9129161B2 Computationally efficient scene classification
  • JP6299427B2 Scene estimation method and scene estimation apparatus
  • US10970561B2 Detecting roadway objects in real-time images
  • US20160207458A1 Real time driving difficulty categorization

Publications

Electroencephalography (EEG), which contains cortical potentials during various mental processes, can be used to provide neural input signals to activate a brain machine interface (BMI). The effectiveness of such an EEG-based prosthetic system would rely on correct classification of executed motor signals from imagined motor movement signals; an executed motor signal should initiate movement in the artificial limb while signals from motor imagery should be filtered out. This work evaluates the performance of features based on average EEG signal power contained in different frequency bands in order to distinguish between the executed and imagined EEG signals. We also investigate Independent Component Analysis (ICA) as a scheme to remove irrelevant artifacts from the EEG signals. Results demonstrate that using EEG for classification can be performed effectively; however results vary significantly from patient to patient, suggesting that BMI must highly specialized for an individual patient.

  • Hierarchical context-aware hand detection algorithm for naturalistic driving 2016 IEEE 19th International Conference on Intelligent Transportation Systems (ITSC), Rio de Janeiro, Brazil, 2016, pp. 1291-1297, doi: 10.1109/ITSC.2016.7795723

    Hand activity is a critical monitoring component in understanding a driver’s behavior within the car. Current vision-based hand detection algorithms perform poorly in naturalistic settings, due to various challenges such as global illumination changes and constant hand deformation and occlusion. To achieve a more accurate and robust hand detection system, this paper presents a hierarchical context-aware hand detection algorithm, which explicitly explores context cues in the vehicle such as prevalent hand shapes and locations, preferred driving habit and coupling effect between multiple hands. The proposed context-aware hand detection algorithm significantly outperforms the state-of-the-art on the VIVA hand dataset.

  • CHASE Algorithm: “Ease of Driving” Classification2015 IEEE 18th International Conference on Intelligent Transportation Systems, Gran Canaria, Spain, 2015, pp. 644-649, doi: 10.1109/ITSC.2015.111

    In this paper, we present the concept of “Ease of Driving” or EoD rating for road scene environment. We introduce the Classification by Holistic Analysis of Scene Environment (CHASE) algorithm to automatically classify the EoD rating using computer vision and machine learning techniques. With the CHASE algorithm we were able to achieve an accuracy of 86.84 % to classify the EoD. In the near future, EoD information will be used to create better navigation services to improve the driving comfort and mobility by leveraging the existing camera sensors in the car.

  • “Ease of Driving” Road Classification for Night-time Driving Conditions SAE World Congress and Exhibition 2016-01-0119, 2016, https://doi.org/10.4271/2016-01-0119

This paper is an extension of our previous work on the CHASE (Classification by Holistic Analysis of Scene Environment) algorithm, that automatically classifies the driving complexity of a road scene image during day-time conditions and assigns it an ‘Ease of Driving’ (EoD) score. At night, apart from traffic variations and road type conditions, illumination changes are a major predominant factor that affect the road visibility and the driving easiness. In order to resolve the problem of analyzing the driving complexity of roads at night, a brightness detection module is incorporated in our end-to-end nighttime EoD system, which computes the ‘brightness factor’ (bright or dark) for that given night-time road scene. The brightness factor along with a multi-level machine learning classifier is then used to classify the EoD score for a night-time road scene. Our end-to-end ‘Night-time EoD system’ is a real-time onboard system implemented and tested on road scene data collected in Japan. We have improved the scope of the CHASE algorithm for computing EoD score for night-time driving conditions by including a brightness detection module.

We demonstrate a two phase classification method, first of individual pixels, then of fixed regions of pixels for scene classification—the task of assigning posteriors that characterize an entire image. This can be realized with a probabilistic graphical model (PGM), without the characteristic segmentation and aggregation tasks characteristic of visual object recognition. Instead the spatial aspects of the reasoning task are determined separately by a segmented partition of the image that is fixed before feature extraction. The partition generates histograms of pixel classifications treated as virtual evidence to the PGM. We implement a sampling method to learn the PGM using virtual evidence. Tests on a provisional dataset show good (+70%) classification accuracy among most all classes.