Development Blog
This is the development blog for the HelVision Threat Detection System. See the About page for an overview of the project.-
Appearance-Based Body Language Detection
Pre-assaultive behavior cues are often communicated with body language, ie, blading, pacing, hand position, head motion, etc. A reliable means of body language detection is sought. The initial scheme is to use an appearance-based model for body language detection. Basically, have a library of canonical desired poses expressed as COCO-style...
-
Tracking Smoothing and HUD Integration
An issue with the person bounding-boxes provided by the detector is the variability of their sizes. This makes for harder de-duplication and re-identification. By creating bounding boxes from pose information, box size variability was greatly reduced and tracking was improved. Creating a center-of-mass box for tracked persons using only the...
-
Object Tracking Extension
Object detection can sometimes be sporadic, leading to gaps in detection across frames. This complicates object tracking. Of specific interest is multi-object tracking (MOT) in online mode, ie, only the current and prior frames are available, with infinite additional frames expected. Batch MOT works on a finite video files. Reviewed...
-
80% Detector Precision Achieved
Having (painstakingly) improved the training/testing data and completed 50 epochs of training, the new object detector scores are as follows meanAP personAP handgunAP rifleAP knifeAP 0.8035 0.8325 0.8521 0.6999 0.8297 AP=average precision Less rifle training data (815 labels vs 6350 person, 1670 handgun, and 1620 knife labels) likely accounts for...
-
Object Detector Training
The Deep High-Resolution Representation Learning for Human Pose Estimation (CVPR 2019) pose estimator is top-down, ie, it takes as part of its input the bounding boxes of persons. It derives its person bounding boxes from PyTorch’s Faster R-CNN ResNet-50 FPN pre-trained model. Since we also want to detect other objects...