Intelligent system
To support integrated sensing and intell₽σ igent systems, Order of Magnitude≠↑ " has developed two inteασlligent hardware systems, a complex m&★odel based on high-performance computing platf ✘orm and a lightweight model ba>★sed on edge AI chips. The fα∞↑ormer realizes high precision computation×☆ at the cost of large volδ∑₹ ume and large power consuφβδmption, while the laσ₩€tter is characterized by miniaturiz₹÷≈"ation and low power consumption.
Based on the above intell↕φ☆igent hardware platform, a large number₩' of intelligent visual algorithms with indepen€ €dent intellectual property rights <÷≤₹have been accumulated for laασ$ rge security and emerg→♦±®ency industries, repres>↕¥ented by conventional algor₩€↕εithms such as vision enhancement, targe'' t detection, stereo vision, n±>≥avigation and positioning, an↑♥<←d customized algorithms for speci÷×fic industries such as gas c ↓loud recognition, fireworks ∞εα®recognition, and temp®"₽erature analysis.
High performance computi→§∞₹ng platform intelligent hardware
[High performance computing platform intelli₩α$↓gent hardware]
Function description
Compatible with NVIDIA Jets★→£on™ AGX, Orin, NX series modφ$βules, pre-installed Ub₹♥✔©untu 18.04 operating syste₹>m, with a maximum of 200T→✘$OPS AI processing capacity, ac♥™ ★cording to different ≥₩applications using super strong lightweight alα>↑uminum alloy materiaλ©₩≈l design, using fan or no fan str₽βucture conductive heat dissipation,£'♠☆ It has the capabili&∞ty of continuous 24 hours of ultra™π>≥-long MTBF stable operation in the field, and&≥ can complete complex on-site edge computing ₩•and management tasks such as multλ☆Ωi-channel camera access, image fusion and stitc≠₹'πhing, image super resolution proc♣ε₩essing, target identificatio•λn and tracking, as well as integrated control manΩ↓agement, display, storage and communication ac®•cording to different field needs.
Characteristic parameter
●Camera access and video output modes: Network ≈© ₩video transmission suppo≥♦rts RTSP/ONVIF/GB28181 and other interf♥β₹aces such as SDI/CVBS f≤§or real-time video stream transmissi ✔↓Ωon.
●Supports up to 8 c₩✘hannels of camera access;
●Support target (people, v±φ∑♣ehicles, ships, etc.) target recognition and ♦§ tracking, support the extraction of rai¶≥lway, highway, river aδnd other target contour, support image super reso>§πlution processing.
Lightweight edge com ≈€♠puting intelligent hardware
[Lightweight edge computing intelligent¥δ≥ hardware]
Function description
Based on various series of AI chips≈$↓ such as Rockchip RV1126/RK1808/RK3588 or AmLog↓☆&ic A311D, lightweight edge computing©∞® intelligent hardware is built. It has↔ the characteristics of small ☆πλ∞size, low power consumption∏π→, easy integration, and supports specific AI fu ¶×¥nctions such as image enhancement, image fusion₽& β and target recognition. Integr ✘ated into monocular or multi-ocular multi-c"∞hannel vision sensors to provide lightweight edg• ∏e computing support.
Characteristic parameter
●Support multiple types of image ↓÷interface, easy integr₩¥ation; Camera access supports MIPI-CSI, BT6♠®δ56/1120, DVP interface, video output sup ports network interface,≥β SDI, AHD/CVBS interface.
●Support image ISP function;
●Integrated lightweight target (≤ people, vehicles, ships, etc.) target re∑♣∑cognition and tracking algorithm.
Intelligent vision series algorithm
[Target recognition and trackingδ≥§←]
Track and recognize the image of pedestr÷Ω§ians, vehicles, ships, obstacγ ♥©les and other specific target♣♠s. It has been applied in rail tr∞ansit, water transportation and other fields, ☆ ≠£and has the advantages of hig÷h detection accuracy and computati&↔∑♦onal efficiency to meet →"the real-time application ♠ ∏requirements.
[Image enhancement]
More clear image and clear fog and rai÷∑n to blur oriented. Image denoising an π ♦d enhancement, improve image clari ∑δty, is conducive to improving↑∞ the detection accuracy of target recognition al ∑ €gorithms. Low cost cameras can alsoγ↓→® present high quality images, improvingφ¶★ the cost performance and compet♣$•itiveness of products. It also improves the visu§ al system's adaptability in harsh scenes and bad ↓×εweather.
[Depth generation]
3D information of target scene is more c"&"onducive to realizing accura÷'te navigation and positioning and im₹¶Ωproving the accuracy of target detection. φ✘3D image data is generated by estimating λφ> the depth information of target€α™ from monocular image data. It ≠≥→¶has the advantages of h¥™γΩigh computing efficiencφε©y and low application co≥←st. Combined with AI depth esti♠✘mation, it can effectively improve the accuracy o<←®☆f 3D image data.
[Visual enhancement]
In order to pursue better performance in ≥✘infrared or multi-spectral fusion imag∑<e display, the monochromatic image of infrared v∞♠₹ision and multi-spectral pseudo-color im ↓Ωage can be converted into←↕♦ RGB image by algorithm, which can improve↓↓€ the visual experience of non-visible image and € broaden the application range of infrared visu★≠≤al products
[Intelligent algorithms for specific industry♦≤ applications]
When it comes to specific scenes such as p γ✔βower, oil and gas, criε€sis and mining, it puts forward some special AI i ♠↓ntelligent requirements for βφ¥the vision system, which nε₹§eeds to combine the ri§ ch sample data of the industry. Fina£♠lly, it presents intelligen&÷¥t recognition algorithms with industry chara<≥cteristics, such as accurate infrare€ ∑d temperature measur¶↔ement based on graphic classification, intellige☆ nt fireworks recognition based on combinationβ↑" of temperature and iεφ mage features, infrared gas c✘£loud recognition and •£semi-quantitative concentr♥★≤<ation labeling, and dangerous event re÷cognition in high-risk operation€¶¶<s.
