TTC.PA.1581 : ENRICHED AND DISCRIMINATIVE CONVOLUTIONAL NEURAL NETWORK FEATURES FOR PEDESTRIAN RE-IDENTIFICATION
PROBLEM ADDRESSED
Pedestrian re-identification and tracking across multiple cameras with a non-overlapped field of view for continuous retrieval of pedestrian walking trajectories in wide areas.
APPLICATION AREAS
- Smart construction
workers’ safety & site analysis - Smart mobility
walkability evaluation - Smart city
urban design & mobility - Smart transportation
traffic flow analysis - Smart retail
behavioural analysis - Smart warehouse
behavioural/ flow analysis
TECHNOLOGY INNOVATIONS
- An approach of explainable CNN design based on feature map visualization
- A new CNN to extract discriminative & distributed features for more robust pedestrian ReID
- An incremental multi-image feature aggregation mechanism for more robust pedestrian identity matching
KEY IMPACTS & ADVANTAGES
Accurate pedestrian tracking with 4-8% improvement over baselines.
Facilitate future research evaluating CNNs for image processing tasks based on an explainable approach.
Enable real-world multi-camera processing toidentify pedestrian walking trajectories.
Support pedestrian walking behavioral analyses and what-if facility layout evaluation and design optimization.
GET IN TOUCH
to receive a personalized consultation, discuss research or commercial collaboration opportunities, and more
info@autosafe.ai
+852 6538 6246