Progress Integrated object detection server with web app and raspi on a crude IRMan mechanical structure. Pipeline work s under conditions with no interrupts. Recorded the time used for calibration: 103 seconds in normal case. Improve data post processing to make sure the position calculated for the same device on Read more…
Progress Completed training Yolov3 model for 4 classes, 200 Epochs in total. Accuracy of the trained model is not yet confirmed, will validate later. Modified CV server to migrate from OpenCV inference to PyTorch inference. Improved style by wrapping up threads in separate classes. Tested the possibility of panorama, reached Read more…
Progress Trained primary yolov3 model supporting one class only (AC). Model being able to recognize AC in camera captured frame. Tested image stitching before object detection. Turned out that object detection fails for panorama, suspecting the reason being the warping edges and irregular input image size. Gave up using panorama Read more…
Progress Improved CV pipeline by supporting multiple message types. Now object detecting server stores requested device list before inference and save inference results of different IrMans separately. Tested pipeline flow from object detector server on AWS EC2 instance to client on RPi. Successfully transmitted image from client to server for Read more…
Progress Added multi-threading feature to the local socket object-detector. Now separates image transfer and computer vision into two different threads, and the gathered image is passed between threads through a shared image queue. Data post-processing is substituted by drawing back, and will be completed as we start to output motor Read more…
Progress Establish local web socket image transfer and object detection pipeline using pertained yolov3 model Tested camera frame gather and object detection on RPi locally, tested with different resolution to see the limit of yolov3 model (Since we would like image transfer to capture less network bandwidth) Start labeling Vizio Read more…
Progress Finished polishing design review documents, add references to corresponding parts Set up AWS EC2 GPU instance for training purposes, configuring environment and pre-trained model Decided to start working on TV from 220 lab, starting to collect data for later labeling Tested LIRC database for 220 TV Schedule Finish setting Read more…
Progress Decided the specific model to use: YoloV3, and researched on ways to train it. Not on campus from Tuesday to Friday Schedule Up till now it seems that I’m on schedule Shall start to integrate software architecture according to the Gantt chart Deliverables Next Week Basic structure of a Read more…
Previously Done With Team (Week 1 to Week 5): Gantt chart planned IR circuits built and worked Camera tested Progress: Made sure LIRC module runs successfully on RPi Planned detailed computer vision pipeline with input output specified Working On: Researching specific computer vision model to use Planned Finish assigned parts Read more…