QNX MonoRange

QNX MonoRange combines a lightweight COCO object detector with a custom multitask neural network trained on KITTI. The detector identifies objects and supplies bounding boxes, while the distance model processes each cropped object to predict: Object class: Car, Pedestrian, or Cyclist Classification confidence Estimated forward distance in metres The distance model learns directly from KITTI 3D location-Z labels using RGB crops. It predicts log-distance and does not use focal length, calibration files, physical object heights, pinhole geometry, or residual correction during inference. The complete pipeline runs natively on QNX 8 and presents live camera frames with object boxes, labels, estimated distances, and FPS. It is an experimental research prototype and may be affected by lighting, camera viewpoint, and domain shift outside KITTI.

Built With

• 5
• c++17
• camera
• cmake
• mobilenet
• opencv
• opengl
• python
• pytorch
• qnx
• raspberry-pi
• sensor
• tensorflow
• torchvision