Automotive Camera [Apply Computer vision, Deep learning] - 1

Explore camera basics for advanced driver assistance systems and autonomous driving, including SAE levels, sensor roles, camera types, OpenCV, Python, and key terms like image sensors and field of view.

Discover how advanced driver assistance systems (ADAS) boost road safety with warnings and limited vehicle control, including adaptive cruise control and lane departure warning and prevention.

Explore the transition from ADAS to autonomous driving, detailing SAE automation levels and how lidar, cameras, and radar enable object detection and tracking in perception, planning, and control.

Explore how exteroceptive and proprioceptive sensors, especially cameras, enable perception in ADAS and autonomous driving, including object and lane detection, sensor fusion, and multi-sensor challenges.

Discover how camera sensors enable real-time object and lane detection with deep learning, and how GNSS and IMU enable sensor fusion for autonomous driving.

Explore camera applications in ADAS and autonomous driving, comparing mono and stereo cameras, depth via disparity, and how calibration and OpenCV enable traffic sign, lane, and object recognition and tracking.

Define ego and target vehicles, map sensor data from multiple sensors to a common ego frame via extrinsic calibration, and review 2d/3d coordinates, yaw, pitch, and roll.

Learn camera fundamentals by examining a camera module, image sensor types (CCD and CMOS), pixels, sensor size, resolution, and how focal length shapes the field of view.

Explore angular field of view, horizontal and vertical field of view, and how sensor height, focal length, and sensor size shape automotive camera performance, frame rate, and interfaces.

Recaps the journey through advanced driver assistance systems, autonomous driving SAE levels, sensor types (radars, camera, lidar, ultrasonic), and camera applications, OpenCV, and ego-vehicle coordination terms.

Explore camera fundamentals, including pinhole model and intrinsic/extrinsic calibration, and survey perception tasks such as classification, localization, detection (R-CNN, YOLO, SSD), segmentation, and multi-object tracking with Kalman filters for ADAS.

Explore the pinhole camera as the foundation of automotive imaging, covering perspective projection, world, camera, image, and pixel coordinate systems, intrinsic and extrinsic matrices, and basic camera hardware.

Explore the central projection model from world to camera coordinates, the image plane at f, and converting to 2d pixel coordinates using the intrinsic matrix and the principal point.

Learn how intrinsic and extrinsic matrices map 3d world coordinates to 2d image pixels, using rotation, translation, homogeneous 4x4 forms, and radial and tangential distortion correction in OpenCV.

Learn how to perform camera calibration, including intrinsic calibration with the K matrix and distortion vector D, and extrinsic calibration with [R|t] using checkerboard targets.

Master intrinsic calibration with a printed checkerboard and ROS or OpenCV tools, and perform extrinsic calibration using world coordinates, targets, and PnP/SVD methods.

Learn how image localization identifies a single object's position in the image plane using bounding boxes, coordinate formats, and landmarks, and distinguish it from object detection.

Discover object detection that combines classification and localization to identify multiple objects and output bounding boxes; compare two-stage detectors with region proposals to one-stage detectors like YOLO and SSD.

Explore the R-CNN family of object detectors, from R-CNN to Fast R-CNN to Faster R-CNN, focusing on regional proposals, ROI pooling, and the regional proposal network.

Explore how YOLO v3 uses a single-stage detector with three output layers to detect objects by predicting bounding boxes and probabilities from a 416 by 416 image, without regional proposals.

Demonstrate how YOLOv3 decodes network outputs into class predictions and bounding boxes using grid cells, anchor boxes, and thresholding, with non-max suppression to finalize detections.

Learn how Yolo v3 decodes multiple bounding boxes into confident detections, uses intersection over union and non-max suppression to suppress duplicates, and finalizes two cars based on a probability threshold.

Learn the single shot detector (SSD) for object detection, compare it with YOLOv3, and explore backbone options like MobileNet, SSD head with multi-scale features, anchor boxes, and non-max suppression.

Explore image segmentation, where every pixel is classified into a category, and distinguish semantic versus instance segmentation, with architectures like U-net and mask R-CNN used in ADAS and autonomous driving.

Enable object detection and instance segmentation with Mask R-CNN, producing bounding boxes and per-instance masks via a faster R-CNN backbone for autonomous driving applications.

Explore how autonomous vehicle perception fuses sensors to detect, classify, and track dynamic objects in the 2d image plane, using bounding boxes, Kalman filters, and deep learning for data association.

Explore how multi object tracking extends single object tracking using bounding box centers and size changes, and how data association and track management handle new, vanishing, and reappearing objects.

Explain multi object tracking through prediction, data association, and measurement update, initialize tracks from bounding boxes, and apply Kalman filter equations to manage detections and vanishing tracks.

Define state representations for multi-object tracking in camera images and apply a constant velocity motion model for prediction, plus data association using Euclidean distance, IOU, and feature matching.

Explore multi-object tracking concepts by detailing the measurement model, prediction and update steps, Kalman filter gain, and track management, mapping eight states to four measurements with H, R, and K.

Build a complete camera perception pipeline for autonomous driving by acquiring images, applying object detection, and tracking multiple objects with a Kalman filter to produce track lists.

Review the full camera perception pipeline from intrinsic and extrinsic calibration to object detection and segmentation, covering AlexNet, YOLO, R-CNN, U-Net, and Kalman-based multi-object tracking.