TI Autonomous Driving Algorithms (TIADALG) Library User Guide
Parking Spot And Vehicle Detection in Fish Eye Images

This model is bsed on MobileNetV1 feature extractor and SSD meta architecure with enhancement for key points detection in fish eye images.

Input

Input to the model is 3 RGB channels of 768x384 image.

Output

Output from the model is list of detected objects. Where each object is represneted by bounding box parameters [top_left_x, top_left_y, bot_right_x, bot_right_y] and 4 kepoints x-y co-ordinates. These four corner points represents four corner points of object enclosed in rectangular bounding box.

Steps for Model importing for TIDL

  1. cd <TIDL_INSTALL_DIR>/ti_dl/utils/tidlModelImport/
  2. ./out/tidl_model_import.out <TIADALG_DIR>/tiadalg_object_detection/utils/tidl_import_jpsdvd.txt

Dependent APIs

  1. tiadalg_image_preprocessing - Raw input data can be formatting for basic scaling, mean subtraction and de-interleaving before feeding to TIDL
  2. tiadalg_fish_eye_transforamtion - Output from TIDL can be brought to rectilinear image domain using this API

Openvx Application

  1. Refer to vision_apps/apps/dl_demos/app_tidl_avp application

Sample Output

Image 1
Image 2

© Copyright 2018 Texas Instruments Incorporated. All rights reserved.
Document generated by doxygen 1.8.6