Object Detection, Tracking and Analytics using Pretrained ML models

Title: Object Detection and Tracking, Posture Detection and Tracking, Animal Gate Tracking with associated variety of analysis ML PreTrained Algorithms: Yolo v5-8, Yolo NAS, DeepSort, DeepLabcut, Detectron 2 Detection and Tracking: Segmentation/Mask and objects Primary Purpose: The goal is to develop a standalone application (windows) to perform ML inference and analytics on variety of objects behaviours in ideally real-time. The interface should enable uploading of images/videos/webcam/stream and perform inference using pretrained weight linked to aforementioned algorithms. The interface should allow simple interaction by the user including drawing a line and/or rectangle on the processed data to collect further analytics such as counting different objects (classes) passing through the defined line/region in different directions (restricted to 4 directions). Realtime (or near realtime) inference and analytics are important for enhanced user experience. Some applications in mind for this tool are traffic monitoring and driver/pedestrian analysis, Sports for players behaviour analysis using detection, tracking and posture analysis, Ergonomics analysis of operators, animal behaviour analysis for scientific studies, etc. Object Detection, Tracking, and Analytics using pretrained AI models such as Yolo v5-8, Yolo-NAS, Deepsort, Detectron 2, deeplabcut etc. All these models are available on Github with pretrained weights. Three key AI functionalities I'm after are 1. object detection, tracking and Analysis 2. Posture detection, tracking and Analytics 3. Animal behaviour monitoring and analysis. The interface should be able to allow basic basic selection of region/area on the input data to collect relevant statistics. The selection could be a line or a rectangular region on the image/video to collect statistics such as number of objects (distinct classes) passed through in either direction, heatmap showing concentration of distinct (selected classes from a list of detected ones) over time, path trajectory of selected classes in different colours to understand the motion behaviour, etc. There are enough examples of all the requirements above and associated codes on the Github. The interface should seamlessly use different ml models on the same input (image, video, webcam, stream) and produce desired statistics. Some of the applications of this tool in mind are traffic behaviour monitoring, sports analytics, human/animal/insects behaviour for scientific studies. The webapp should also generate a high level simple report of the statistics estimated during analysis. - Desired Level of Accuracy: High accuracy with some false positives - ML Model Preference: The client has models in mind that are required Skills and Experience: - Experience in implementing object detection, tracking, and analytics using pretrained ML models such as Yolo V5-8, Yolo_NAS, detectron 2, Deepsort and deeplabcut. - Strong understanding of computer vision and machine learning algorithms - Proficiency in Python and relevant ML libraries/frameworks (e.g., TensorFlow, PyTorch) - Ability to fine-tune and optimize ML models for desired accuracy and performance - Experience in working with video data and real-time processing - Knowledge of retail analytics and customer behaviour analysis is a plus - Experience with React.js for front end development - Experience with WebRTC for realtime data transfer of videos/streams for enhanced useability