Perception accelerated.
Perciv.ai is a spin-off from TU Delft's Intelligent Vehicles Group focused on making AI-driven machine perception affordable for all.
We provide expertise in multi-sensor integration, sensor fusion and AI-driven perception solutions for autonomous driving, UAVs, smart cities and more.
Our Expertise
From sensor selection to the downstream perception task performance analysis, Perciv.ai handles it all.
Next-gen Sensors
Imaging Radars (4D), Solid-state/Doppler LiDARs
For sensor manufacturers:
We sample and investigate next-generation machine perception sensor prototypes from companies worldwide.
Our sensor evaluation services help accelerate prototype development and demonstrate sensor capabilities for manufacturers.
For sensor users:
Different sensors like vision, LiDAR, and 4D imaging radar offer unique capabilities. A well-informed decision is critical for both the efficacy and affordability of your perception solution.
Based on our experience with various LiDAR, camera, and radar sensors, we can help select and evaluate the best sensors for any use case.
Sensor Suite Integration
Synchronization, Calibration
Accurate calibration and synchronization of sensors are critical for enabling fusion techniques, as we have learned from recording and publishing multiple automotive datasets.
We developed a multi-sensor suite capable of integrating new automotive sensors (and prototypes) and have experience designing custom sensor suites, such as for UAVs. We are open to new challenges.
High-Quality Data Recording
Multi-modal, Annotated
Our team has years of experience creating high-quality automotive datasets for reliable and trustworthy perception solution development, such as the View-of-Delft and Eurocity datasets.
We can record high-quality datasets with selected sensors and prototypes in a variety of scenarios, including urban, rural, and highway.
We offer help with 2D and 3D annotation, either manual or semi-automated.
Perception Algorithms
Deep Learning, State-of-the-art
Our team has conducted state-of-the-art research in machine perception for multiple fields, including automotive, surveillance, and UAVs.
We have extensive knowledge on the topic of automotive radars, including research on radar point clouds and custom-designed neural networks for range-doppler data.
Perciv AI can help apply cutting-edge algorithms or design custom methods (single sensor or sensor fusion) to showcase our customers' sensors or solve their challenges efficiently.
-
CNN Based Road User Detection Using the 3D Radar Cube
-
Occlusion-aware sensor fusion for early crossing pedestrian detection.
-
Multi-class Road User Detection with 3+1D Radar in the View-of-Delft Dataset
-
3+1D Radar and Camera based Point Cloud Estimation for 3D Road User Detection
Task-Oriented Performance Analysis
Qualitative, Quantitative
We provide precise, thorough results through meaningful, task-oriented metrics for end perception tasks, as well as a qualitative demonstration of sensor capabilities.
We can share insights on performance capabilities and practical feasibility to help sensor manufacturers and technology vendors reduce collaboration and product development turn-around time.
Vulnerable road user detection in an urban scenario (Delft city centre) by fusing a 4D radar and a monocular camera.
Partners
News
