Of course, to can according to demand the local tyrant company of own and custom-built sensor, this piece appear redundant, please circuitous. This piece basically apply to drive automatically primary developer, basically face the researcher of college and place of scientific research courtyard, give priority to with function, consider cost issue lesser. The author basically is based on Linux system to undertake developing, the API of place introduction all is with Ubuntu14.04 and above version accurate (gave Ubuntu18.04 April, 2 years one stabilizes version, be worth to expect) .

 

Graph 1 drive automatically mainstream sensor

 

Only item camera

 

Head turn AVT industry watch for a chance, domestic agent is more. Included camera sort is relatively all ready, interface includes interface of 1394 front line, net mouth interface to wait, the price differs to tens of thousands of yuan from thousands of yuan.

 

Application: The application development of only item camera basically includes a feature kind of the symbol detect with identify, if driveway line detects, light of traffic sign identifying, traffic identifies, pedestrian and car detect etc, the visual computation that is based on machine study is driving automatically popular day can be essential part certainly, although eye comes round to say the vision detects,dependability is not very tall, drive automatically in what be main perception method with lidar the application in car was not achieved anticipate.

 

Graph 2 road surface and car identify

 

Binocular camera

 

Apply at pair of eye visions of the setting outdoor not to see more really, the Bumblebee that has used before the author is binocular also be to apply at indoor setting, ZED camera was being contacted later also is indoor better with (after all base line length was secured, so short! ) . The proposal is used at outdoor do visual speedometer to perhaps identify kind of algorithm, build by oneself binocular sensor, below the circumstance that makes sure synchronism sparks, decide base line length according to specific requirement. Binocular the vision does not circle parallactic graph and double target to decide, at present more current double target decides a way is to use Zhang Zhengyou method, use Camera Calibration Toolbox to undertake demarcate, of course, in OpenCV compositive also this method, those who be used to C process designing is OK look, openCV classical teaching material ” study OpenCV ” Chinese edition 464 pages have detailed content to 492 pages, have code more, do not say more.

 

Graph demarcate of 3 binocular and stereo visions

 

Application: It is OK that fraise detects, be confined to is indoor, structural light, close quarters; Visual speedometer is the applied domain that has compared at present really, use at crossing fixed position to wait, all sorts of SLAM opening a source also have application in the technology, be like ORB SLAM2, ” visual SLAM14 is told ” have detailed introduction and application, good book, recommend.

 

Panoramic camera

 

Cent is single shot panoramic camera and joining together of much camera lens are panoramic camera. The development application that the author is based on panoramic vision is not much (use when much ocular head, develop its not completely panoramic advantage) , one has seen their car is making up seemingly when Tecent made vista map 2015.

 

Application: Recommend quite do visual speedometer with panoramic camera, ken is big, diagnostic dot correlation is spent tall, the individual feels to cooperate to combine navigation to do high accuracy map to rebuild with panoramic vision from beginning to end is a right choice, implementation drives automatically fixed position of car driveway level, do not have thorough research otherly.

 

Infra-red camera

 

Infra-red camera should attribute another kind of the vision, night inspecting the effect is better than daytime effect, applicable detects at pedestrian, car etc. After feeling to appear in lidar previously, infra-red camera is driving automatically the position that one is in to compare awkwardness in application, the price does not poor and what come without lidar result is direct, after using, discover, in fraise (like the person) identify on, laser disperse chooses the successive picture on the cloud or space of be not a patch on. Additional, the amazed discovery in the test, body of infra-red photograph machines and tools can catch the people on glass! Fine think of extremely fear! Infra-red camera can is opposite on certain level calorific body undertakes distinguishing, if road surface, pedestrian waits, but need an aftertreatment after all, use absolutely height without lidar or gradient undertakes fraise detects those who come is direct. Chromatic camera can be replaced below night condition, fraise is inspected to detect before undertaking with monitoring.

 

Graph 4 infra-red visions are become picture

 

Millimeter wave radar

 

Apply to high speed environment and form into columns travel, other setting was not necessary to install seemingly. Chan Maigui, wholesale and cheap. A when had used Delphi, also should be domestic application is more, on the distance farthermost go to 150 meters, cent dogs for only place and much target detection works two kinds mode, former apply to form into columns the car before travel process is medium detects, what latter applies to high speed environment to fall is remote target detection. As a result of millimeter wave radar into become for cone like the principle picture (be equivalent to from outside disperse the internal heat with sudorifics becomes a cone) , the area of the fraise in basis cone comes computative fraise, consequently to complex setting, fraise is relatively mixed and disorderly. And freeway setting is relatively ideal, road environment is good, car is opposite less, apply to ahead car shift to an earlier date early-warning. (Tesla thinks vision + millimeter is enough prop up drive automatically system, the individual comparatives self-identity, but still need to advance one step on algorithm seemingly)

 

Development: Send and receive data through Can, data format is relatively simple, there is a definition in general agreement file, not panicky also without the foundation to Can mouth development, quite simple, buy an USBCAN converter, what the Can below Ubuntu system receives data depend on a library, the data later is analytic come according to the agreement can. Lidar.

 

Application: Fraise detects the mainstream, …… of manufacturer of Waymo, Baidu, each old traditional car, each big poineering company, each big college, academy place is so much drive automatically steam roof is making up it is lidar almost (Tesla except) ; High accuracy map rebuilds to build a model with the environment; SLAM.

 

Graph much radar is shirt-sleeve

 

The sensor that the author in be being developed actually at present uses with respect to so much, hope to be helped somewhat to abecedarian. It is OK also to have those who choose difficult disease to equipment type selecting illicit letter I, great communicating.