A5103136068- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Autonomous Vehicle Technology and Safety
- Cognitive Computing and Networks
- Water Quality Monitoring Technologies
- Advanced Image and Video Retrieval Techniques
- Robotics and Sensor-Based Localization
- Neural dynamics and brain function
- Industrial Vision Systems and Defect Detection
- Ferroelectric and Negative Capacitance Devices
- Advanced Memory and Neural Computing
- Human Pose and Action Recognition
- Image Retrieval and Classification Techniques
- Advanced Vision and Imaging
- Image Processing Techniques and Applications
- Fire Detection and Safety Systems
Harbin Institute of Technology
2019-2022
Monocular 3D object detection, with the aim of predicting geometric properties on-road objects, is a promising research topic for intelligent perception systems autonomous driving. Most state-of-the-art methods follow keypoint-based paradigm, where keypoints objects are predicted and employed as basis regressing other properties. In this work, unified network named FADNet presented to address task monocular detection. contrast previous methods, we propose divide output modalities into...
Online multi-object tracking (MOT) is an active research topic in the domain of computer vision. Although many previously proposed algorithms have exhibited decent results, issue tracklet inactivation has not been sufficiently studied. Simple strategies such as using a fixed threshold on classification scores are adopted, yielding undesirable mistakes and limiting overall performance. In this paper, conditional random field (CRF) based framework put forward to tackle online MOT problems. A...
An essential element for intelligent perception in mechatronic and robotic systems (M&RS) is the visual object detection algorithm. With ever-increasing advance of artificial neural networks (ANN), researchers have proposed numerous ANN-based methods that proven to be effective. However, with cumbersome structures do not befit real-time scenarios M&RS, necessitating techniques model compression. In paper, a novel approach training light-weight developed by revisiting knowledge...
Monocular 3D object detection is a prevalent research topic with several issues yet to be explored. In this paper, we present novel framework for monocular autonomous driving as the target application. As first contribution, wed esign Convolution-Involution Hybrid module regression of geometric properties. The peculiar characteristics involutions are delicately combined mature design convolutions yield enhanced network performance. second contribution us Dynamic Error- Tolerant loss. A...
We present a human tumble action recognition method based on Spiking Neuron Network. The approach adopts which can convert skeleton sequence into 2D image by equating the corresponding X, Y, Z components of joint coordinates to R, G, B pixels. After conversion, final images are fed recognized Network (SNN) model is mechanisms with increased biological plausibility instead being subsequently converted rate-based network. SNN architecture uses leaky-integrate-and-fire (LIF) neurons,...