loading

OVERVIEW

Since 2015, Vispect has been committing on research of AI-based object recognition algorithms and multi-sensor fusion.

We are passionate to apply the latest Computer Vision and Artificial Intelligence technology to vehicles and machinery, to improve driving and operating safety through new products.

Until January 2024, Vispect has obtained 47 intellectual property rights, including six patents and 41 software copyrights, while most of these properties are about artificial intelligence and automatic control application.


24
25

COMPUTER VISION & AI

With cameras as the main sensor, our AI-based algorithms detect, from the camera images, vehicle, pedestrian, cyclist, road lane line, traffic sign, driver face, and driver gesture. Detected objects are tracked, and algorithms keep analyzing their trends and ‘intentions’, finally, actions, like alerting the drivers or even stopping the machines, are to be taken.

From devices on vehicles , to the cloud-based platform, to the Vispect dedicated statistical and analytical software, we have built a virtuous circle to improve the algorithms. Footage and images by 4G wireless from the in-service vehicles (only under permission of the fleet owner and the driver. Faces obscured by algorithm and without personal information, according to the local laws) provide raw materials for next level machine-learning and accuracy analysis, thus the accuracy of Vis detection, by means of both True and False Positive, has been greatly improved to a considerable high level. In a in-service project, the below statistic table of 41,334 pieces of alarm videos indicates the False Positive detection rates of a Vis G4-S system with four top-down cameras were less than 0.5%, and this means the True Positive rates exceeded 99.5%.

24
24
Above: Vispect dedicated statistical and analytical software is a powerful tool to count and analyze the outcomes of detection algorithms.

24
25

HARDWARE DEVELOPMENT​

Our experienced Hardware Development team has been designing a variety of devices and systems as the solid base to run our AI algorithms.
In-house development includes planning, performance evaluation, schematic design, protection design, industrial standard adaption, circuit design, PCB layouting, engineering prototype, and New Product Introduction, until the Massive Production stage. Designs are to meet EMC, EMI, and environmental standards of ISO and EN, or meet special requirements for car maker OEM fitting.
Multi-core SOC is adopted as the brain of Vispect system, while some models even have more than one SOC to speed up AI computing. Until now, we have used SOC from ADT, Qualcomm, MTK, RK, and etc. MCU are used as the controlling core of the systems. Until now, we have used MCU from ST, NXP, Atmel, Microchip, and etc.


24
25

SOFTWARE ENGINEERING​

A group of around 20 experienced engineers forms the Vispect software team. In-house works include analyzing and planning, firmware development, algorithm development, software development, software test, road test, and Verification Test. Android and Linux play as the Operating Systems of most Vispect SOCs. V-Model is strictly followed through the whole software development process. White box software tests are applied on key or safety-related modules, while black-box tests applied on other modules. For high-level safety systems, Safety Integrity Level 2 (SIL2) is being followed.


24
25

LiDAR

LiDAR, as the secondary sensor, performs excellently in Vispect systems.
We use both single-plane LiDAR and multi-plane LiDAR as well. Processed by our algorithms, LiDAR cloud points are clustered for object detection, or fused with camera pixels for advanced procession.
As camera has more pixels than LiDAR cloud points, camera is doing better on object classification, while because of its nature, LiDar is bringing us more benefits on various measuring, such as Distance Measuring, Height Detection, or Deflection Measurement. .


AUTOMATIC CONTROL​

Vispect systems have interfaces to exchange information with, to interact with, and thus to control the vehicles and machines. Various new functions can be deployed based on.
This feature, for example, to stop a heavy Articulated Dump Truck when the Vispect system finds a pedestrian too close in front of the truck, or to brake a forklift when a worker walks in its approaching direction.


24
25

COMMUNICATION​

To exchange information with other systems, general Vispect systems provide a diverse range of interfaces, including CAN, 485, 232, Ethernet and T1, USB, video out, GPIO to receive and send ON/OFF signals, and 4G wireless to cloud-based data center.
Interfaces are usually developed according to OEM requirements. Most vehicles and machines use CAN interfaces. In some cases, to enhance safety level, combined interfaces have be used. For example, an Automatic Train Protection system is using two Ethernet interfaces to exchange data with a Vispect system, the mutual backup solution. Another example, a OEM forklift is using one high/low voltage wire and a CAN interface to interact with a Vispect unit.


24
25
24
24

OTHER SENSORS

Some Vispect systems have sensors other than camera and LiDAR. Gyro sensor is adopted to detect acceleration, angular velocity sensor to detect movement on a plane, ultrasonic sensor to detect nearby obstacle at low speed, temperature sensor to detect overheat at point, thermal image sensor to find abnormal temperature in space.


24
25

OTA & REMOTE OPERATION

Some Vispect models support Over-The-Air upgrading. Software of the SOCs in a Vispect Host can be updated without people going to the site. The Host also supports remote setting and remote maintenance. This means you don't need to go to the vehicle to modify the running parameters of a Vispect device, instead, you just need to log in your on-line account to input the new values. For European applications, R156 regulation must be met for remote upgrading and setting.


24
25

DATA SECURITY

In some cases, photos and video footage are to be stored in Vispect systems, or even to be sent to cloud-based data centers through 4G wireless. For European applications, these photos and videos are protected according to the R155 regulation and GDPR norms. The users have the rights to decide whether to encrypt these photos and videos or not.