Systems Software Engineer

Stellenbeschreibung:

Systems Software Engineer

Work Location: Munich
Experience: 4-5 years
Contract duration: 12 months
Working arrangement: 3 days per week in the Munich office, 2 days remote

Role Summary

Responsible for designing, implementing, and validating a camera-based lane detection and tracking system for an Advanced Driver Assistance System (ADAS). The role spans the full pipeline: image preprocessing, neural-network-based lane/marking detection, temporal tracking, lane geometry modeling, and vehicle-state fusion for warning logic.

Key Responsibilities

Analyze requirements and define the system architecture for a front-camera-based lane detection function (inputs, outputs, latency/accuracy targets, ODD — operational design domain).

Design and train a neural network (e.g., segmentation-based, anchor-based, or row-classification-based architectures such as LaneNet, SCNN, UFLD, PolyLaneNet, or transformer-based approaches) for lane marking/lane boundary detection.

Implement lane tracking across frames (Kalman filter, particle filter, or learned temporal models) to ensure stable, jitter-free lane estimates and handle occlusion, worn markings, or missing lanes.

Fit and maintain a lane geometry model (e.g., clothoid/polynomial curve fitting) and estimate vehicle position/heading relative to the lane.

Develop the Lane Departure Warning logic: time-to-lane-crossing (TTLC) estimation, threshold logic, driver intent filtering (e.g., turn signal suppression), and warning triggering strategy.

Integrate camera calibration (intrinsic/extrinsic) and perspective transformation (IPM – inverse perspective mapping) into the pipeline.

Optimize models for embedded/automotive-grade hardware (quantization, pruning, TensorRT/embedded inference frameworks) to meet real-time constraints.

Build datasets, define annotation guidelines, and drive data collection strategy for diverse conditions (rain, night, glare, worn markings, construction zones, curves).

Validate against relevant standards (e.g., Euro NCAP LDW/LKA test protocols) and define test/validation KPIs (false positive/negative rates, detection range, curvature accuracy).

Collaborate with vehicle integration teams; support HIL/vehicle-level testing.

Required Skills & Experience

Strong background in computer vision and deep learning, especially semantic segmentation, keypoint detection, or curve-fitting-based lane detection architectures.

Proficiency in Python and deep learning frameworks (PyTorch).

Solid understanding of classical CV techniques: camera calibration, homography/IPM, edge detection, Hough transforms — useful for hybrid approaches and sanity baselines.