Product introduction:
IAS100 AI Algorithm Video Analysis Software (hereinafter referred to as IAS100) is independently developed by HopeVision. The platform adopts intelligent analysis algorithm based on traditional machine learning + deep learning fusion, which can efficiently realize the detection of things of interest such as people, vehicles, ships, and unmanned vehicles. The structured information extraction of objects such as aircraft and animals can provide effective intelligent analysis results for application software systems, and provide intelligent means for 24-hour unattended monitoring scenarios, saving manpower and material resources. The platform adopts Socket+Http+RestAPI to realize intelligent analysis of detection, classification, tracking and other objects of interest in incoming video frames, realizes intelligent analysis of visible light and thermal image dual-light video, and outputs detection results with structured information.
Traditional algorithms mainly involve background modeling, alarm intrusion, single-target tracking, multi-target trajectory analysis and panoramic stitching. In terms of deep learning, through the training of common targets such as water, land, and air, it has been able to detect and identify people, vehicles, and ships. , drones, and more than 45 targets and actions. Through the application and tempering in various scenarios such as border defense lines, military bases, deserts, lakes, and oceans, the platform has strong environmental adaptability and can effectively cope with various environments with weak vision such as rain, snow, wind, sand, and smoke. .
Based on the advantages in the field of optoelectronic technology, the company provides long-distance target observation and sighting optoelectronic equipment. At the same time, it has the algorithm detection and recognition technology for long-distance small targets in video image processing, and combines the photoelectric control servo control technology to effectively use it in border and coastal defense. , base, chemical industry, forestry and other industries, and have carried out in-depth cultivation and refinement of each field, so that the algorithm technology can be more suitable for the actual use of the scene, and play a huge value. Compared with other manufacturers, our algorithm technology tends to detect and recognize small targets in visible light and thermal imaging, and can do in-depth optimization based on the scene, which has a leading advantage in the same industry.
Features:
1) Enhanced detection of moving targets: Supports the use of target recognition and target trajectory tracking to strengthen the motion detection, line mixing detection and regional intrusion detection of very large and very small targets;
2) Human body action diversity detection: support the detection of the maximum and minimum objects under various movements of the human body, such as squatting, walking, running, jumping, crawling, and swimming;
3) Multiple types of target detection: support the detection of various targets such as people, cars, ships, animals, birds, drones, airborne objects, aircraft, and landing gear;
4) Dual-light video detection: support the detection methods of visible light, thermal imaging video synchronization with the same rules, synchronization with different rules, wrong time with the same rules and wrong time with different rules;
5) Long-distance small targets: support intrusion alarm detection for small targets below 5*5 pixels; support for detection and identification of ultra-small targets below 15*15 pixels;
6) Multi-form deployment of cloud, edge and terminal: support the overall system solution, realize the business integration of software + algorithm; support the cooperation mode of back-end algorithm analysis platform, intelligent algorithm analysis suite and algorithm model authorization.
Application scenarios:
It is widely used in the detection of people and vehicles, the detection of fireworks in forests, oil depots, chemical plants, and parks, the detection of low-altitude drones in airports, nuclear power plants, and cities, and the detection of ports, docks, lakes, and marine ships.
Model name | IAS100 AI algorithm video analysis service platform | |
Applicable equipment | Photoelectric turntable, photoelectric cloud table, electronic sentry, Universal security camera | |
System performance | A single service supports at least 16 channels of real-time video detection (the number of channels is limited by hardware configuration) A single service supports not less than 100 channels of video time-sharing detection (the number of channels is limited by hardware configuration) | |
Image analysis | 1. Target tracking: support single target and multi-target tracking algorithm, real-time display of target trajectory; It supports automatic tracking of target identification (selection) , automatic tracking of target invasion area, automatic tracking of target collision line, manual selection of target tracking, automatic tracking of target algorithm identification (selection) , automatic doubling of tracking process, automatic recovery of target loss; It supports target tracking in water, land and air scenes, and supports KCF and optical flow tracking algorithm Support visible video tracking, thermal imaging video tracking, turntable equipment angular speed tracking and PID tracking, cloud equipment speed level tracking 2. Panoramic mosaics: support for visible light panoramic mosaics, thermal imaging panoramic mosaics; support level 360 ° , pitch ≥50 ° seamless panoramic mosaics; support panoramic image upload and download function 3. Semantic segmentation: visual video semantic segmentation, thermal imaging video semantic segmentation, semantic segmentation based on deeplabv3 series, object segmentation model based on DNANet; Support for people, cars, ships, uavs, animals and other objects, pixel-level segmentation; support the separation of the sky and the ground, small objects and clutter background separation 4. Moving Target Detection: support moving target detection, moving target detection, regional intrusion detection; It can strengthen the motion detection, line-mixing detection and regional intrusion detection of the maximum and minimum targets by means of target identification (matching) and target track tracking Support visible video moving target detection, thermal imaging video moving target detection, real-time display of moving target trajectory | |
Ability to detect moving objects | Visible Light at 1080P resolution supports detection of objects no smaller than 5 × 5 pixels | |
Thermal imaging at D 1 resolution supports detection of objects not less than 3 * 3 pixels | ||
After the enhancement, the detection accuracy is over 95% | ||
Ground target recognition service (optional) | 1. Personnel Detection: support visible light video, thermal video detection of personnel targets; support the human body in squatting, walking, running, jumping, crawling, swimming and other actions under the great and small target detection 2. Support visual video, thermal video on the face detection 3. Vehicle Detection: support visual video, thermal video on the detection of vehicle targets; Support detection of cars, vans, buses, trucks, pick-ups, oil tankers, tricycles, bicycles, excavators, tractors, construction vehicles and other types of vehicle objectives 4. Animal Detection: support visual and thermal video detection of animal targets; support detection of cattle, sheep, horses, camels, cats, dogs and other animals targets 5. Firework Detection: support visible light video detection of smoke and fire targets, thermal imaging video detection of fire; support the sun, reflection, chimney and other common heat source targets detection | |
Detection capability | Visible light can be detected at 1080P resolution The minimum target pixel is not greater than 10 × 10, and the maximum target pixel is not less than 300 × 240 | |
Thermal imaging can be detected at D1 resolution The minimum target pixel is not greater than 8 * 8, and the maximum target pixel is not less than 200 * 160 | ||
When human, vehicle and animal target pixels are higher than 30 × 30, the detection accuracy can reach more than 96% | ||
When the pixel of pyrotechnic target is higher than 50 × 50, the detection accuracy can reach more than 90% | ||
A single video card supports a detection speed of no less than 100 frames/s Single service supports simultaneous detection of no less than 4 graphics cards | ||
Surface target identification service (optional) | 1. Ship Detection: support visual and thermal video detection of ship targets; support detection of passenger ships, cargo ships, fishing boats, speedboats, ships, sand dredgers and other types of ship targets 2. It supports the detection of human body swimming target on the water surface by thermal imaging video 3. Support visual video, thermal video on the detection of fishing target (optional) 4. Support visual video, thermal video on the detection of floating objects (optional) | |
Detection capability | Visible light can be detected at 1080P resolution The minimum target pixel is not greater than 10 × 10, and the maximum target pixel is not less than 300 × 240 | |
Thermal imaging can be detected at D1 resolution The minimum target pixel is not greater than 15 × 15, and the maximum target pixel is not less than 200 × 160 | ||
When the ship target pixel is higher than 30 × 30, the detection accuracy can reach more than 95% | ||
When the target element of swimming and fishing is higher than 35 × 35, the detection accuracy can reach more than 85% | ||
When the target element of floating object is higher than 50 × 50, the detection accuracy can reach more than 85% | ||
A single video card supports a detection speed of no less than 100 frames/s Single service supports simultaneous detection of no less than 4 graphics cards | ||
Low altitude target identification service (optional) | 1. UAV detection: support visual video, thermal video detection of UAV targets; support detection of four-rotor, six-rotor, fixed-wing and other low-altitude UAV targets 2. Detection of airborne objects: support the detection of visible light video on the kite, balloon and other airborne objects; support the detection of thermal imaging video on the balloon 3. Support visible light video, thermal imaging video for low-altitude bird detection 4. Aircraft Detection: support visual video, thermal video detection of aircraft targets; support helicopters, civil aviation, landing gear, fighters (optional) and other targets detection | |
Detection capability | Visible light can be detected at 1080P resolution The minimum target pixel is not greater than 10 × 10, and the maximum target pixel is not less than 300 × 240 | |
Thermal imaging can be detected at D1 resolution The minimum target pixel is not greater than 8 * 8, and the maximum target pixel is not less than 200 * 160 | ||
When the target pixel of UAV is higher than 30 × 30, the detection accuracy can reach more than 95% | ||
When the pixel of the object is higher than 30 × 30, the detection accuracy can reach more than 80% | ||
A single video card supports a detection speed of no less than 100 frames/s Single service supports simultaneous detection of no less than 4 graphics cards | ||
Video Processing | 1. Stream-pulling decoding: video decoding that supports device direct connection and streaming media forwarding and streaming; video decoding, clipping and compression of common video resolution such as 1080P, 1280 * 1024,720P, D1,640 * 512, etc. Support one decoder for multi-model, multi-algorithm synchronization 2. Video Coding: Fusion coding that supports decoding video and reasoning information; compression of encoded video, video service distribution and directional push 3. Reasoning playback: support video frame model reasoning, target marking, occlusion, attribute extraction, target information storage, reasoning video storage and target warning push; Support 1 frame/NS -25 frame/s inferential frame rate control; support inferential record, video query, playback and download 4. Model Management: Support Model Upload, DELETE, upgrade and other management operations; support to add, Delete, browse, rename, hide and other reasoning tag management operations; Support multi-process, multi-model algorithm traffic scheduling; support manual upload video, picture test algorithm model reasoning effect | |
Hardware adaptation | 1. Support for Hays Chips, Atlas, Cambrian, Nvidia and other chips, GPU algorithm transplantation and adaptation 2. Support for cloud, edge and end-to-end form deployment; support for back-end services distributed networking, Edge Smart 1 + N detection and end-to-side small model migration adaptation | |
Interface Protocol | 1. Access Protocol: RTSP, GB28181(optional) video protocol access; support haikang, Dahua, Yushi and other mainstream manufacturers equipment access protocol 2. Secondary development: support SDK secondary development; support Restful API platform docking; support to provide RTSP format reasoning video stream |