2021 ideal one based on horizon journey ® Chip 3 successfully mass produced the world’s first NOA navigation assisted driving scheme equipped with 8mp (megapixel) forward-looking camera. The first release is the most popular model, and the ideal one continues to occupy the first place in the sales list of high-end intelligent models of domestic independent brands. The on-board 8mp front view has immediately become the standard configuration of high-end intelligent models and an important purchase standard for consumers.
Based on the mature mass production application of 8mp front view camera, the powerful l2+ ADAS will no longer be exclusive for high-end models. The mid-range model, even the entry-level model that focuses on cost performance, can also achieve excellent intelligent function experience by carrying the 8mp front view scheme, so as to enhance the competitiveness of the whole vehicle products and help the brand improve.
8mp foresight perception with more “foresight” and “insight”
With the continuous improvement of the intelligent level of vehicles, the driving assistance scheme also puts forward higher requirements for the forward-looking perception ability, and higher resolution cameras have become an inevitable application trend. As a relatively leading and mature visual sensor in the industry, the core value of 8mp vehicle mounted camera for ADAS applications naturally lies in longer detection distance, clearer imaging effect and wider FOV field angle.
Seeing farther means seeing distant objects earlier. For example, if the effective detection distance for vehicles, traffic signs and other long-distance road targets exceeds 250m, it can make correct judgments in advance and smoothly complete the corresponding avoidance behavior, so as to bring users a more smooth and safe auxiliary driving experience.
8mp HD camera will provide more high-quality image data for environmental perception to achieve more accurate visual perception. This has also become one of the best ways to solve the intelligent driving corner case, especially in challenging scenes such as small target detection and dealing with glare changes.
In addition, compared with the industry’s scheme of binocular or trinocular low pixel forward-looking combination, 8mp can achieve more than 120 ° field coverage by virtue of its ultra wide field of view angle, and can identify dangerous scenes such as “⻤ probe” and “lateral cut in” earlier, greatly improving the lateral safety performance of the vehicle. At the same time, monocular assembly will greatly reduce the difficulty of vehicle layout, and the shape is relatively more beautiful. It is a highly cost-effective scheme.
8mp powerful forward-looking perception
With the combination of strong and strong, horizon launched a forward-looking perception scheme suitable for 8mp
The mass production of 8mp means that the on-board AI chip needs to have enough computing power to support the perception calculation of a larger amount of data. At the same time, it also needs to have a matching perception algorithm, especially in the stability and accuracy of long-distance detection, lateral vehicle cut in detection and vehicle ranging and speed measurement.
Horizon Matrix ® Mono3 is the horizon based journey ® 3 vehicle specification AI chip, monocular forward-looking perception scheme for L2 and above ADAS markets. On the basis of mono2’s algorithm verified in large-scale production, mono3 can process multiple AI tasks more efficiently and flexibly, and realize real-time detection and accurate recognition by adapting an 8million ultra-high pixel forward-looking camera.
Horizon journey 3 vehicle specification AI chip
Mono3 can fully call the computing power of journey 3 with high performance, low power consumption and low delay by giving full play to the advantages of software and hardware synergy, and transform it into a high-level output of key technology performance such as perception;
The multi task network framework designed in combination with the characteristics of journey chip can provide rich, stable and accurate perceptual output and support more function expansion;
Aiming at the special optimization in the localization scene in China, the high-precision detection and effective recognition of “no one has me” are realized on the difficult detection targets such as Chinese characteristic vehicle models, lane lines and traffic signs;
Through the advanced monocular depth estimation technology, by analyzing the disparity changes of local pixels in the front and back frames, we can judge whether there is an obstacle in front, and then make reasonable avoidance or braking action, fully expand the perceptual boundary and improve security;
Based on the real 3D and freespace ranging scheme, it can effectively solve the ranging across vehicles and reduce the security problems caused by missing target detection.
Mono 3 differentiated technology innovation for visually challenging scenes
Mono3 key algorithm indicators catch up with and surpass international head players
Mono3 can be found in [email protected] Under the low delay performance of, effective perception supports more than 10 categories of dynamic target perception and more than 100 categories of static target perception. At the same time, it supports 16 categories of road targets to expand semantic segmentation. The detection rate for key areas is as high as 99.8%.
Better perceptual performance, farther, wider and clearer
Depending on the longer detection distance, mono3 scheme can support 250m long-distance moving vehicle tracking detection (including 3D), support more than 180m static target detection (such as cone barrel and construction area), support 100m long-distance pedestrian and two wheeled vehicle detection (including 3D), greatly improve the detection ability of key road elements and lane lines at intersections, and support 100m traffic sign and traffic light detection.
Based on the wider lateral perception range, vehicles in adjacent lanes appear in the forward horizontal view of the vehicle. Even in the case of occlusion, mono3 can still detect cut in vehicles earlier. The lane line detection ability has been greatly improved, and it can flexibly respond to changes in the number and structure of lanes, especially for large curvature curves, with better perceptual performance.
With clearer imaging effect, mono3 can support the detection of cone barrels, construction signs, triangle warning signs, diversion signs and other special road obstacles; It supports the detection of lane lines in and out of the tunnel and in the tunnel under light changes.
More powerful visual expansion capability, supporting high-precision positioning and mapping
Mono3 can supplement lane line recognition results by means of re projection and high-precision map to achieve Lane level positioning accuracy. By acquiring the information of complex driving scenes (toll stations, intersections, large curvature curves, etc.), a priori and over distance perception can be realized, and the corresponding relationship between lanes and traffic elements (traffic lights, signs, road arrows) can be provided for association matching, so as to achieve more accurate road condition perception.
Mono3 can support real-time scene modeling on the vehicle end, output the reconstructed local 3D semantic map, and provide customers with cloud map aggregation service reference design; At the same time, based on the detection of triangle warning signs, cone barrels and other roadblocks, the collection and reporting of dynamic traffic information such as lane level accident restricted areas and construction areas can be realized, significantly improving the real-time and accuracy of road condition navigation.
More powerful auxiliary driving ability, more safe and comfortable driving experience
With the enhanced perception ability of 8mp HD camera, mono3 can identify vehicle targets and VRU targets earlier on the basis of L2 function (acc/tjp), avoid behaviors in advance, and ensure the safety of assisted driving. At the same time, under the premise of high coverage of intersections, large curvature curves, tunnels and other scenes, it can achieve a lower artificial takeover rate and bring users a more comfortable and smooth driving experience.
Based on the powerful forward-looking perception ability of mono3, the integration of other sensors and high-precision maps is expanded through the form of domain controller. It can also support complex high-speed l2+ NOA navigation assistant driving functions such as lane level navigation, automatic lane change and overtaking, automatic ramp up and down, etc. The support of mono3 visual perception for positioning and mapping functions will also bring higher quality products and services to users at a more affordable cost.
Mono3 helps accelerate the popularization of high-level assisted driving
Mono3 is the only ultra-low power solution in the market that can support the form of 8mp forward-looking all-in-one machine. With the extreme energy efficiency performance of mono3 under the combination of software and hardware, it will break through the performance ceiling of the front-view all-in-one machine, and the moped enterprises will realize higher configuration and higher-level intelligent driving functions in a shorter cycle, lower cost and smaller changes. The large-scale mass production application of mono3 for the full-scale intelligent models will also greatly improve the popularity of ADAS, thus driving the acceleration of the intelligent process of the entire automobile industry.
Market scale forecast for future 8mp foresight
Horizon mono series is the only domestic monocular vision ADAS scheme that has achieved mass production of front loading at present. It has won more than 10 fixed points of models in total. Mono3 has also become the world’s first mass production 8mp front vision perception scheme. Facing the large-scale implementation of ADAS, horizon insists on positioning tier-2. By providing an open platform for journey chips, it continues to make collaborative innovation with industrial partners in a flexible and open cooperation mode, supporting customers to create differentiated products and creating a better intelligent driving experience for consumers.