
Machine Learning (ML) and Artificial intelligence (AI) for Object Classification
AWARE for On-Road & ANTHEM for Off-Road Perception
AWARE and ANTHEM are based on Special Neural Network training process with core Development and Milestones consist of:
Training and Test Data creation
Using Sensor Fusion System to create training data Manual refinement of training data
Using the full RDM
Identification of location and object classification at the same time
Using Preprocessing steps
Improved accuracy by introduction of edge detection algorithm
Following Hybrid approach
Further improvement by implementing a hybrid approach using identified subsets of the RDM
Improving the results by incorporating meta information as inputs
Additional data like target angle, distance and velocity lead to further improvements
Reality Check – Optimization Phase
Optimization is needed to fit the algorithms into the data flow of the system and the physical limitations.
Accuracy
-
Reached 94% accuracy goal after optimizing for throughput by considerably reducing the algorithm complexity and manual labeling percentage
-
Pedestrians
-
Vehicles
-
Bicycles & motorcycles
-
Static objects
-
Complexity
-
Adjust to limitation of available memory
-
Adapting to radar RDM
Throughput
-
Limit the data required to identify objects
-
Identification speed keeps up with frame rate
Learn more about our
Real-Time Perception AI
Send us an email: