Research Engineer (CV/ML)

Position Summary

We’re building safety first video telematics products (ADAS/DMS/driver behavior analytics) that run efficiently on edge devices inside commercial vehicles. You will write modern C++ software, integrate and optimize CV/ML pipelines, and ship reliable, low latency perception features such as driver monitoring and distance estimation from camera feeds.

Key Responsibilities

· Own C++ software modules for on device video capture, preprocessing, inference, and post processing on Linux.

· Implement classical image processing pipelines (denoise, resize, color space, undistortion) and CV algorithms (keypoints, homography, optical flow, tracking).

· Build and optimize distance/spacing estimation from monocular/stereo camera(s) using calibration, geometry, and/or depth‑estimation networks.

· Integrate ML models (PyTorch/TensorFlow → ONNX/TensorRT/NNAPI/NPU runtimes) for DMS/ADAS events: drowsiness, distraction/gaze, phone‑usage, smoking, seat belt, etc.

· Hit real time targets (FPS/latency/memory) on CPU/GPU/NPU using SIMD/NEON, multithreading, zero copy buffers.

· Write clean, testable C++, CMake builds, and Git based workflows (branching, PRs, code reviews, CI).

· Instrument logging/telemetry; debug with gdb/addr2line, sanitize and profile with perf/valgrind.

· Collaborate with data/ML teams on dataset curation, labeling specs, training/evaluation, and model handoff.

· Work with product & compliance to meet on road reliability, privacy, and regulatory expectations.

Qualifications

· B.Tech/B.E. in CS/EE/ECE (or equivalent practical experience).

· 2–3 years in CV/ML or video‑centric software roles. Hands on in modern C++ on Linux, with strong Git and CMake .

· Solid image processing and computer‑vision foundations (camera models, intrinsics/extrinsics, distortion, PnP, epipolar geometry).

· Practical experience integrating CV/ML models on device (OpenCV + ONNX Runtime/TensorRT/NCNN/MediaPipe/NNAPI).

· Experience building real time pipelines for live video (GStreamer/FFmpeg, RTSP/RTMP, ring buffers), optimizing for latency & memory .

· Competence in multithreading/concurrency , lock free queues, and producer–consumer designs.

· Comfort with debugging & profiling on Linux targets.

Reporting To: Technical Lead ADAS

Requisites:

· Experience with driver monitoring or ADAS features; event logic and thresholding for production alerts.

· Knowledge of monocular depth estimation, stereo matching, or structure from motion for distance estimation .

· Model training exposure ( PyTorch/TensorFlow ): augmentation, evaluation (precision/recall, ROC/PR), quantization/pruning, conversion to ONNX/TensorRT/NCNN.

· Hardware acceleration (GPU/VPU/NPU, Arm NEON /DSP), YOLO/RT DETR/Lightweight backbones on edge.

· Cross compiling, Yocto/Buildroot, containerized toolchains; unit tests (gtest), static analysis (clang tidy, cppcheck), sanitizers.

· Basic familiarity with MQTT/IoT , message schemas, and over the air updates.

Technical Competency:

· Languages: C++, Python

· CV/ML: OpenCV, ONNX Runtime/TensorRT/NCNN/MediaPipe; PyTorch/TensorFlow (for training/eval).

· Video: GStreamer/FFmpeg, V4L2, RTSP/RTMP.

· Build/DevOps: CMake, Git, gtest, clang‑tidy, sanitizers; CI/CD (GitHub/GitLab/Bitbucket).

· Debug/Perf: gdb, perf, valgrind